Search

With the vigorous development of intelligence agriculture, the progress of automated large-scale and intensive pig farming has accelerated significantly. As a biological feature, the pig face has important research significance for precise breeding of pigs and traceability of health. In the management of live pigs, many managers adopt traditional methods, including color marking and RFID identification, but there will be problems such as off-label, mixed-label and waste of manpower. This work proposes a non-invasive way to study the identification of multiple individuals in pigs. The model was to first replace the original backbone network of YOLOv4 with MobileNet-v3, a popular lightweight network. Then depth-wise separable convolution was adopted in YOLOv4′s feature extraction network SPP and PANet to further reduce network parameters. Moreover, CBAM attention mechanism formed by the concatenation of CAM and SAM was added to PANet to ensure the network accuracy while reducing the model weight. The introduction of multi-attention mechanism selectively strengthened key areas of pig face and filtered out weak correlation features, so as to improve the overall model effect. Finally, an improved MobileNetv3-YOLOv4-PACNet (M-YOLOv4-C) network model was proposed to identify individual sows. The mAP were 98.15 %, the detection speed FPS were 106.3frames/s, and the model parameter size was only 44.74 MB, which can be well implanted into the small-volume pig house management sensors and applied to the pig management system in a lightweight, fast and accurate manner. This model will provide model support for subsequent pig behavior recognition and posture analysis.

Abstract Objective To analyze primary angle closure suspect (PACS) patients’ anatomical characteristics of anterior chamber configuration, and to establish artificial intelligence (AI)-aided diagnostic system for PACS screening. Methods A total of 1668 scans of 839 patients were included in this cross-sectional study. The subjects were divided into two groups: PACS group and normal group. With anterior segment optical coherence tomography scans, the anatomical diversity between two groups was compared, and anterior segment structure features of PACS were extracted. Then, AI-aided diagnostic system was constructed, which based different algorithms such as classification and regression tree (CART), random forest (RF), logistic regression (LR), VGG-16 and Alexnet. Then the diagnostic efficiencies of different algorithms were evaluated, and compared with junior physicians and experienced ophthalmologists. Results RF [sensitivity (Se) = 0.84; specificity (Sp) = 0.92; positive predict value (PPV) = 0.82; negative predict value (NPV) = 0.95; area under the curve (AUC) = 0.90] and CART (Se = 0.76, Sp = 0.93, PPV = 0.85, NPV = 0.92, AUC = 0.90) showed better performance than LR (Se = 0.68, Sp = 0.91, PPV = 0.79, NPV = 0.90, AUC = 0.86). In convolutional neural networks (CNN), Alexnet (Se = 0.83, Sp = 0.95, PPV = 0.92, NPV = 0.87, AUC = 0.85) was better than VGG-16 (Se = 0.84, Sp = 0.90, PPV = 0.85, NPV = 0.90, AUC = 0.79). The performance of 2 CNN algorithms was better than 5 junior physicians, and the mean value of diagnostic indicators of 2 CNN algorithm was similar to experienced ophthalmologists. Conclusion PACS patients have distinct anatomical characteristics compared with health controls. AI models for PACS screening are reliable and powerful, equivalent to experienced ophthalmologists.

Abstract Osteoporosis, a silent yet debilitating disease, presents a significant challenge due to its asymptomatic nature until fractures occur. Rapid bone loss outpaces regeneration, leading to pain, disability, and loss of independence. Early detection is pivotal for effective management and fracture risk reduction, yet current diagnostic methods are time-consuming. Despite its importance, research addressing early diagnosis remains limited. Deep learning, particularly convolutional neural networks (CNNs), has emerged as a potent tool in image analysis. This paper presents a novel approach utilizing transfer learning with CNNs for osteoporosis detection from X-ray images. The proposed approach not only achieves a high accuracy of osteoporosis diagnosis but also offers a revealed feature map that can guide medical professionals for osteoporosis diagnosis. The innovation lies in a dual strategy: (i) a model integrating transfer learning from CNN architectures such as AlexNet, VGG-16, ResNet-50, VGG-19, InceptionNet, XceptionNet, and a custom CNN, and (ii) a dataset collection augmentation mechanism to enhance learning accuracy. The study includes binary and multiclass classification of knee joint X-ray images into normal, osteopenia, and osteoporosis groups, utilizing a dataset of 1947 knee X-rays for training and testing. Performance comparisons against state-of-the-art models reveal the proposed VGG-19 model achieves the highest accuracy at 92.0% for multiclass and 97.5% for binary. These findings underscore the potential of deep learning with transfer learning in aiding early osteoporosis detection, thereby mitigating fracture risks.

In radiation therapy, breathing and other influences cause a constant movement of the tissue to be irradiated. Thus, a continuous position control is required which could be handled by the usage of 3D ultrasound imaging. For this purpose, two approaches for target tracking in 3D ultrasound (US) sequences of the liver are analyzed in this study. Therefore, an image-by-image localization of the target is performed using a deep localization network. A singletarget and a multiple-target approach are investigated where deep localization networks are trained for locating one specific and multiple specific targets, respectively. Training the networks and evaluating the tracking algorithm is performed on the basis of a labeled 3D US liver data set in 2-fold crossvalidation experiments. The single-target and multiple-target approaches performed comparable with a mean tracking error of 2.28±1.20mm and 2.23±1.28 mm, respectively. The proposed tracking algorithm is real-time capable with a mean runtime per 3D ultrasound image of 68ms.

PURPOSE: Spontaneous intracerebral hemorrhage (ICH) is the most severe form of stroke. The timely assessment of early hematoma enlargement and its proper treatment are of great significance in curbing the deterioration and improving the prognosis of patients with ICH. This study aimed to develop an automated hybrid approach to predict hematoma expansion in ICH. METHODS: The transfer learning method was applied to build a hybrid model based on a convolutional neural network (CNN) to predict the expansion of hematoma. The model integrated (1) a CNN for automated hematoma segmentation and (2) a CNN-based classifier for hematoma expansion prediction that incorporated both 2-dimensional images and the radiomics features of the 3-dimensional hematoma shape. RESULTS: The radiomics feature module had the highest area under the receiver operating characteristic curve (AUC) of 0.58, a precision of 0, a recall of 0, and an average precision (AP) of 0.26. The ResNet50 and Inception_v3 modules had AUCs of 0.79 and 0.93, a precision of 0.56 and 0.86, a recall of 0.42 and 0.75, and an AP of 0.51 and 0.85, respectively. Radiomic with Inception_v3 and Radiomic with ResNet50 had AUCs of 0.95 and 0.81, a precision of 0.90 and 0.57, a recall of 0.79 and 0.17, and an AP of 0.87 and 0.69, respectively. CONCLUSION: A model using deep learning and radiomics was successfully developed. This model can reliably predict the hematoma expansion of ICH with a fully automated process based on non-contrast computed tomography imaging. Furthermore, the radiomics fusion with the Inception_v3 model had the highest accuracy.

Abstract Mosquito-borne diseases are a major global health threat. Traditional morphological or molecular methods for identifying mosquito species often require specialized expertise or expensive laboratory equipment. The use of convolutional neural networks (CNNs) to identify mosquito species based on images may offer a promising alternative, but their practical implementation often remains limited. This study explores the applicability of CNNs in classifying mosquito species. It compares the efficacy of body and wing depictions across three image collection methods: a smartphone, macro-lens attached to a smartphone and a professional stereomicroscope. The study included 796 specimens of four morphologically similar Aedes species, Aedes aegypti, Ae. albopictus, Ae. koreicus and Ae. japonicus japonicus. The findings of this study indicate that CNN models demonstrate superior performance in wing-based classification 87.6% (95% CI: 84.2–91.0) compared to body-based classification 78.9% (95% CI: 77.7–80.0). Nevertheless, there are notable limitations of CNNs as they perform reliably across multiple devices only when trained specifically on those devices, resulting in an average decline of mean accuracy by 14%, even with extensive image augmentation. Additionally, we also estimate the required training data volume for effective classification, noting a reduced requirement for wing-based classification compared to body-based methods. Our study underscores the viability of both body and wing classification methods for mosquito species identification while emphasizing the need to address practical constraints in developing accessible classification systems. Graphical abstract

Abstract Skin conditions are becoming increasingly prevalent across the world in current times. With the rise in dermatological disorders, there is a need for computerized techniques that are completely noninvasive to patients’ skin. As a result, deep learning models have become standard for the computerized detection of skin diseases. The performance efficiency of these models improves with access to more data with their primary aim being image classification. In this paper, we present a skin disease detection methodology using image processing techniques, non-local means denoising and convolutional neural network (CNN) backed by sparse dictionary learning. Here, the major benefit of using NLM denoising followed by sparse dictionary learning with CNNs in image classification lies in leveraging a multi-stage approach that enhances the quality of input data, extracts meaningful and discriminative features, and improves the overall performance of the classification model. This combined approach addresses challenges such as noise robustness, feature extraction, and classification accuracy, making it particularly effective in complex image analysis tasks. For denoising, the average Peak Signal to Noise Ratio (PSNR) obtained for images from HAM-10000 dataset is 33.59 dB. For the ISIC-2019 dataset, the average PSNR for the train folder is 34.37 dB, and for the test folder it is 34.39 dB. The deep learning network is trained for the analysis of skin cancer images using a CNN model and is achieving acceptable results in classifying skin cancer types. The datasets used contain high-resolution images. After all the tests, the accuracy obtained is 85.61% for the HAM-10000 dataset and 81.23% for the ISIC-2019 dataset, which is on par with existing approaches validated by benchmarking results.

Abstract To meet the needs of automated medical analysis of brain tumor magnetic resonance imaging, this study introduces an enhanced instance segmentation method built upon mask region-based convolutional neural network. By incorporating squeeze-and-excitation networks, a channel attention mechanism, and concatenated attention neural network, a spatial attention mechanism, the model can more adeptly focus on the critical regions and finer details of brain tumors. Residual network-50 combined attention module and feature pyramid network as the backbone network to effectively capture multi-scale characteristics of brain tumors. At the same time, the region proposal network and region of interest align technology were used to ensure that the segmentation area matched the actual tumor morphology. The originality of the research lies in the deep residual network that combines attention mechanism with feature pyramid network to replace the backbone based on mask region convolutional neural network, achieving an improvement in the efficiency of brain tumor feature extraction. After a series of experiments, the precision of the model is 90.72%, which is 0.76% higher than that of the original model. Recall was 91.68%, an increase of 0.95%; Mean Intersection over Union was 94.56%, an increase of 1.39%. This method achieves precise segmentation of brain tumor magnetic resonance imaging, and doctors can easily and accurately locate the tumor area through the segmentation results, thereby quickly measuring the diameter, area, and other information of the tumor, providing doctors with more comprehensive diagnostic information.

Abstract The multi-lead electrocardiogram (ECG) is widely utilized in clinical diagnosis and monitoring of cardiac conditions. The advancement of deep learning has led to the emergence of automated multi-lead ECG diagnostic networks, which have become essential in the fields of biomedical engineering and clinical cardiac disease diagnosis. Intelligent ECG diagnosis techniques encompass Recurrent Neural Networks (RNN), Transformers, and Convolutional Neural Networks (CNN). While CNN is capable of extracting local spatial information from images, it lacks the ability to learn global spatial features and temporal memory features. Conversely, RNN relies on time and can retain significant sequential features. However, they are not proficient in extracting lengthy dependencies of sequence data in practical scenarios. The self-attention mechanism in the Transformer model has the capability of global feature extraction, but it does not adequately prioritize local features and cannot extract spatial and channel features. This paper proposes STFAC-ECGNet, a model that incorporates CAMV-RNN block, CBMV-CNN block, and TSEF block to enhance the performance of the model by integrating the strengths of CNN, RNN, and Transformer. The CAMV-RNN block incorporates a coordinated adaptive simplified self-attention module that adaptively carries out global sequence feature retention and enhances spatial–temporal information. The CBMV-CNN block integrates spatial and channel attentional mechanism modules in a skip connection, enabling the fusion of spatial and channel information. The TSEF block implements enhanced multi-scale fusion of image spatial and sequence temporal features. In this study, comprehensive experiments were conducted using the PTB-XL large publicly available ECG dataset and the China Physiological Signal Challenge 2018 (CPSC2018) database. The results indicate that STFAC-ECGNet surpasses other cutting-edge techniques in multiple tasks, showcasing robustness and generalization.

Abstract To improve students’ ability to recognize and appreciate artworks, and further enhance their academic performance and classroom satisfaction, this study explores the application of the Convolutional Neural Network (CNN) model based on optimization in art teaching. Firstly, the importance and challenges of art teaching are analyzed. Secondly, the principle and structure of CNN and its application in the classification field are expounded, and then the CNN classification model is optimized. Finally, the effectiveness of the optimized model is verified by experiments. Experimental results show that the optimized model’s accuracy is up to 95.2% in the performance evaluation. The training time of the optimized model is much lower than that of the traditional model, and this model still maintains 95.2% accuracy under the noise of 14.7%. In addition, the accuracy of the optimized model on the unseen test data is 92%. In comparing teaching experiment results, by introducing the CNN classification model, Class B students’ average score of art homework has increased by 4.3 points. The score for class satisfaction is 8.1 points. This indicates that the optimized CNN model has significant advantages in art teaching and can effectively improve students’ classroom satisfaction and academic performance. Therefore, this study has specific reference significance for the innovation of the art teaching model.

Abstract The global interest in market prediction has driven the adoption of advanced technologies beyond traditional statistical models. This paper explores the use of machine learning and deep learning techniques for stock market forecasting. We propose a comprehensive approach that includes efficient feature selection, data preprocessing, and classification methodologies. The wavelet transform method is employed for data cleaning and noise reduction. Feature selection is optimized using the Dandelion Optimization Algorithm (DOA), identifying the most relevant input features. A novel hybrid model, 3D-CNN-GRU, integrating a 3D convolutional neural network with a gated recurrent unit, is developed for stock market data analysis. Hyperparameter tuning is facilitated by the Blood Coagulation Algorithm (BCA), enhancing model performance. Our methodology achieves a remarkable prediction accuracy of 99.14%, demonstrating robustness and efficacy in stock market forecasting applications. While our model shows significant promise, it is limited by the scope of the dataset, which includes only the Nifty 50 index. Broader implications of this work suggest that incorporating additional datasets and exploring different market scenarios could further validate and enhance the model's applicability. Future research could focus on implementing this approach in varied financial contexts to ensure robustness and generalizability.

Abstract Background Recently, the process of evolution information and the deep learning network has promoted the improvement of protein contact prediction methods. Nevertheless, still remain some bottleneck: (1) One of the bottlenecks is the prediction of orphans and other fewer evolution information proteins. (2) The other bottleneck is the method of predicting single-sequence-based proteins mainly focuses on selecting protein sequence features and tuning the neural network architecture, However, while the deeper neural networks improve prediction accuracy, there is still the problem of increasing the computational burden. Compared with other neural networks in the field of protein prediction, the graph neural network has the following advantages: due to the advantage of revealing the topology structure via graph neural network and being able to take advantage of the hierarchical structure and local connectivity of graph neural networks has certain advantages in capturing the features of different levels of abstraction in protein molecules. When using protein sequence and structure information for joint training, the dependencies between the two kinds of information can be better captured. And it can process protein molecular structures of different lengths and shapes, while traditional neural networks need to convert proteins into fixed-size vectors or matrices for processing. Results Here, we propose a single-sequence-based protein contact map predictor PCP-GC-LM, with dual-level graph neural networks and convolution networks. Our method performs better with other single-sequence-based predictors in different independent tests. In addition, to verify the validity of our method against complex protein structures, we will also compare it with other methods in two homodimers protein test sets (DeepHomo test dataset and CASP-CAPRI target dataset). Furthermore, we also perform ablation experiments to demonstrate the necessity of a dual graph network. In all, our framework presents new modules to accurately predict inter-chain contact maps in protein and it’s also useful to analyze interactions in other types of protein complexes.

Abstract Many image fusion methods have been proposed to leverage the advantages of functional and anatomical images while compensating for their shortcomings. These methods integrate functional and anatomical images while presenting physiological and metabolic organ information, making their diagnostic efficiency far greater than that of single-modal images. Currently, most existing multimodal medical imaging fusion methods are based on multiscale transformation, which involves obtaining pyramid features through multiscale transformation. Low-resolution images are used to analyse approximate image features, and high-resolution images are used to analyse detailed image features. Different fusion rules are applied to achieve feature fusion at different scales. Although these fusion methods based on multiscale transformation can effectively achieve multimodal medical image fusion, much detailed information is lost during multiscale and inverse transformation, resulting in blurred edges and a loss of detail in the fusion images. A multimodal medical image fusion method based on interval gradients and convolutional neural networks is proposed to overcome this problem. First, this method uses interval gradients for image decomposition to obtain structure and texture images. Second, deep neural networks are used to extract perception images. Three methods are used to fuse structure, texture, and perception images. Last, the images are combined to obtain the final fusion image after colour transformation. Compared with the reference algorithms, the proposed method performs better in multiple objective indicators of $$Q_{EN}$$ Q EN , $$Q_{NIQE}$$ Q NIQE , $$Q_{SD}$$ Q SD , $$Q_{SSEQ}$$ Q SSEQ and $$Q_{TMQI}$$ Q TMQI .

Objective Prevention and protection of foreign objects intrusion into railway becomes a hot research topic, and the division of intrusion areas requires detection of the track location. In order to determine the track location in the image, a method based on improved Mask R-CNN (Mask region convolutional neural network) combined with mathematical model is proposed to identify and extract the railway track. Method Firstly, Mask R-CNN is optimized and added with attention mechanism, and transfer learning is introduced into the above method to improve the model′s generalization ability. Then the optimized model is used to identify and segment the track, extracting the segmentation data and fitting with the corresponding mathematical expression to realize the identification and extraction of the track. The track in the image is divided into four categories and labeled with the Labelme labeling software, forming a training set and a test set. The training set is used to train the optimized model, and the test set to evaluate the detection results of the optimized model. Result & Conclusion The research results show that compared with other segmenting models and the original model, the proposed method performs well under the same training intensity, reaching 97.5% of the accuracy rate in track type determination, and basically above 80% in track segmentation. Tests show that the proposed method improves the general applicability of the detection by using the good performance of neural network, and can accurately determine the track type and segment the track.

This research endeavors to address the persistent dissemination of public opinion within social networks, mitigate the propagation of inappropriate content on these platforms, and enhance the overall service quality of social networks. To achieve these objectives, Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) techniques are employed in this research to develop a predictive model for anticipating public opinion crises in social network mining. This model furnishes users with a valuable reference for subsequent decisionmaking processes. The initial phase of this research involves the collection of user behavior data from social networks using IoT technologies, serving as the basis for extensive big data analysis and neural network research. Subsequently, a social network text categorization model is constructed by amalgamating the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture, elucidating the training procedures of deep learning methodologies within CNN and LSTM networks. The effectiveness of this approach is subsequently validated through comparisons with other deep learning techniques. Based on the obtained results and findings, the CNN-LSTM model demonstrates a noteworthy accuracy rate of 92.19% and an exceptionally low loss value of 0.4075. Of particular significance is the classification accuracy of the CNN-LSTM algorithm within social network datasets, which surpasses that of alternative algorithms, including CNN (by 6.31%), LSTM (by 4.43%), RNN (by 3.51%), Transformer (by 40.29%), and Generative Adversarial Network (GAN) (by 4.49%). This underscores the effectiveness of the CNN-LSTM algorithm in the realm of social network text classification.

This paper is dedicated to exploring the practical implementation of deep learning and Internet of Things (IoT) technology within systems designed for recognizing human motion behavior. It places a particular emphasis on evaluating performance in complex environments, aiming to mitigate challenges such as poor robustness and high computational workload encountered in conventional human motion behavior recognition approaches by employing Convolutional Neural Networks (CNN). The primary focus is on enhancing the performance of human motion behavior recognition systems for real-world scenarios, optimizing them for real-time accuracy, and enhancing their suitability for practical applications. Specifically, the paper investigates human motion behavior recognition using CNN, where the parameters of the CNN model are fine-tuned to improve recognition performance. The paper commences by delineating the process and methodology employed for human motion recognition, followed by an in-depth exploration of the CNN model's application in recognizing human motion behavior. To acquire data depicting human motion behavior in authentic settings, the Internet of Things (IoT) is utilized for extracting relevant information from the living environment. The dataset chosen for human motion behavior recognition is the Royal Institute of Technology (KTH) database. The analysis demonstrates that the network training loss function reaches a minimum value of 0.0001. Leveraging the trained CNN model, the recognition accuracy for human motion behavior achieves peak performance, registering an average accuracy of 94.41%. Notably, the recognition accuracy for static motion behavior generally exceeds that for dynamic motion behavior across different models. The CNN-based human motion behavior recognition method exhibits promising results in both static and dynamic behavior recognition scenarios. Furthermore, the paper advocates for the use of IoT in collecting human motion behavior data in real-world living environments, contributing to the advancement of human motion behavior recognition technology and its application in diverse domains such as intelligent surveillance and health management. The research findings carry significant implications for furthering the development of human motion behavior recognition technology and enhancing its applications in areas such as intelligent surveillance and health management.

In recent years, deep learning-based object detection algorithms have been a hot topic in computer vision research, with the YOLO (you only look once) algorithm standing out as an excellent object detection algorithm. The evolution of its network architecture has played a crucial role in improving detection speed and accuracy. This paper conducts a comprehensive horizontal analysis of the overall frameworks of YOLOv1 to YOLOv9, comparing the network architecture (backbone network, neck layers and head layers) and loss functions. The strengths and limitations of different improvement methods are thoroughly discussed, with a specific evaluation of the impact of these improvements on model accuracy. This paper also delves into discussions on dataset selection and construction methods, the rationale behind choosing different evaluation metrics, and their applicability and limitations in various application scenarios. It further explores specific improvement methods for YOLO algorithm in five application domains (industrial, transportation, remote sensing, agriculture, biology), and discusses the balance among detection speed, accuracy, and complexity in these application domains. Finally, this paper analyzes the current development status of YOLO in various fields, summarizes existing issues in YOLO algorithm research through specific examples, and in conjunction with the trends in application domains, provides an outlook on the future of the YOLO algorithm. It also offers detailed explanations for four future research directions of YOLO (multi-task learning, edge computing, multimodal integration, virtual and augmented reality technology).

Reactor core transient calculation is very important for the reactor safety analysis, in which the kernel is neutron kinetics calculation by simulating the variation of neutron density or thermal power over time. Compared with the point kinetics method, the time-space neutron kinetics calculation can provide accurate variation of neutron density in both space and time domain. But it consumes a lot of resources. It is necessary to develop a surrogate model that can quickly obtain the temporal and spatial variation information of neutron density or power with acceptable calculation accuracy. This paper uses the time-varying characteristics of power to construct a time function, parameterizes the time-varying characteristics which contains the information about the spatial change of power. Thereby, the amount of targets to predict in the space domain is compressed. A surrogate method using the machine learning is proposed in this paper. In the construction of a neural network, the input is processed by a convolutional layer, followed by a fully connected layer or a deconvolution layer. For the problem of time sequence disturbance, a structure combining convolutional neural network and recurrent neural network is used. It is verified in the tests of a series of 1D, 2D and 3D reactor models. The predicted values obtained using the constructed neural network models in these tests are in good agreement with the reference values, showing the powerful potential of the surrogate models.

Modern approaches with advanced technology can automate and expand the extent and resolution of biodiversity monitoring. We present the development of an innovative system for automated wildlife monitoring in a coastal Natura 2000 nature reserve of the Netherlands with 65 wireless 4G wildlife cameras which are deployed autonomously in the field with 12 V/2A solar panels, i.e. without the need to replace batteries or manually retrieve SD cards. The cameras transmit images automatically (through a mobile network) to a sensor portal, which contains a PostgreSQL database and functionalities for automated task scheduling and data management, allowing scientists and site managers via a web interface to view images and remotely monitor sensor performance (e.g. number of uploaded files, battery status and SD card storage of cameras). The camera trap sampling design combines a grid-based sampling stratified by major habitats with the camera placement along a traditional monitoring route, and with an experimental set-up inside and outside large herbivore exclosures. This provides opportunities for studying the distribution, habitat use, activity, phenology, population structure and community composition of wildlife species and allows comparison of traditional with novel monitoring approaches. Images are transferred via application programming interfaces to external services for automated species identification and long-term data storage. A deep learning model for species identification was tested and showed promising results for identifying focal species. Furthermore, a detailed cost analysis revealed that establishment costs of the automated system are higher but the annual operating costs much lower than those for traditional camera trapping, resulting in the automated system being >40 % more cost-efficient. The developed end-to-end data pipeline demonstrates that continuous monitoring with automated wildlife camera networks is feasible and cost-efficient, with multiple benefits for extending the current monitoring methods. The system can be applied in open habitats of other nature reserves with mobile network coverage.

Switchgear is a vital component of modern power systems, responsible for regulating the flow of electrical power. Contact issues, irregular loads, and other similar problems can cause switchgear to overheat, leading to unexpected disturbances and potential damage to the power equipment. Thermal imaging shows significant potential and is increasingly employed to detect faults in power equipment. However, the unique characteristics of thermal images often pose challenges to accurate fault detection. This research aims to study the effectiveness of transfer learning architectures for switchgear fault detection. This paper applies eleven transfer learning architectures, including SqueezeNet, GoogLeNet, InceptionV3, DenseNet201, ResNet50, Xception, InceptionResNetV2, ShuffleNet, EfficientNetB0, AlexNet, and VGG19. The results of the testing phase demonstrated that the application of transfer learning by fine-tuning pre-trained convolutional neural network architectures was highly effective in the classification of thermal images captured from switchgear units. The models achieved accuracy rates between 83.87% and 98.38%, and values of F1-Scores between 83.11% and 98.34% in the pre-trained architectures.

Dijital dönüşüm ile tedarik zinciri yönetiminde büyük veri analitiğinin önemi gün geçtikçe artmaktadır. Özellikle müşteri taleplerinin hızlı ve doğru tahmin edilmesinde büyük verinin kullanımı firmalara rekabet avantajı sağlamaktadır. Bu doğrultuda, yapay zekâ tekniklerinden biri olan derin öğrenme modelleri büyük verideki karmaşık örüntülerin keşfedilmesinde öne çıkmaktadır. Son yıllarda literatürde çok sayıda derin öğrenme yöntemi önerilmiştir. Bu çalışmada, satış tahmini problemi için derin öğrenme yöntemlerinin performansları karşılaştırılmıştır. Bu kapsamda derin sinir ağı (DNN), derin otokodlayıcı (Deep AE), evrişimli sinir ağı (CNN), tekrarlayan sinir ağı (RNN), uzun kısa-süreli bellek (LSTM) ağı, çift yönlü LSTM (Bi-LSTM) ağı, kapılı tekrarlayan birim (GRU), CNN-LSTM ve evrişimli LSTM (ConvLSTM) yöntemleri uygulanmıştır. Çeşitli sektörlere ait satış verileri kullanılarak deneysel çalışmalar gerçekleştirilmiştir. Hiperparametre optimizasyonu ardından ele alınan yöntemlerin performansları tahmin doğruluğu ve eğitim süreleri açısından karşılaştırılarak sonuçların istatistiksel anlamlılığı değerlendirilmiştir. Sonuç olarak, LSTM ve GRU modellerinin tahmin doğruluğunda başarılı sonuçlar verdiği, CNN modelinin ise eğitim süresini kısalttığı görülmüştür.

The article proposes a method for calculating features in an image to form a training data set for solving various problems of video image classification. This method involves the use of well-known integral transformations — the Radon transform and the Steklov function. The proposed method is compared with convolutional neural networks both in terms of the percentage of correct prediction and in terms of its execution time. As a test task, the problem of finding a fragment of a photograph containing an image of a car is considered.

Abstract Background Seed vigor identification is critical to guaranteeing the quality and yield of maize. Although seeds with impaired vigor may germinate under normal conditions, planting under unfavorable conditions makes it difficult to produce healthy plants. Therefore, non-destructive and rapid detection of seed vigor using hyperspectral imaging (HSI) technology is crucial for improving crop production efficiency. Methods Hyperspectral images of maize seeds were acquired employing the HSI system, the original spectra were preprocessed using Savitzky–Golay smoothing and multiplicative scatter correction, and the feature wavelengths were extracted using the successive projections algorithm (SPA). Discriminant models were constructed based on support vector machine (SVM), random forest, artificial neural network (ANN), and convolutional neural network (CNN-DC). Results The results showed that SVM, ANN, and CNN-DC could discriminate well between maize seeds with different vigor levels, and their accuracy rate was over 70%. The SPA algorithm showed that the RMSE value achieved a minimum of 0.3406, while the number of variables was 49. The CNN-DC model outperformed the other models, which reached the highest accuracy of 92.06%. This study demonstrates that DL combined with HSI has excellent potential for identifying seed vigor. Conclusions This study shows that the proposed method has excellent results for hyperspectral image data processing and can accurately identify maize seed vigor.

Abstract Epithelial ovarian cancer (EOC) is widely recognized as the most lethal gynecological malignancy; however, its early-stage detection remains a considerable clinical challenge. To address this, we have introduced a new method, named Comprehensive Serum Glycopeptide Spectral Analysis (CSGSA), which detects early-stage cancer by combining glycan alterations in serum glycoproteins with tumor markers. We detected 1712 glycopeptides using liquid chromatography–mass spectrometry from the sera obtained from 564 patients with EOC and 1149 controls across 13 institutions. Furthermore, we used a convolutional neural network to analyze the expression patterns of the glycopeptides and tumor markers. Using this approach, we successfully differentiated early-stage EOC (Stage I) from non-EOC, with an area under the curve (AUC) of 0.924 in receiver operating characteristic (ROC) analysis. This method markedly outperforms conventional tumor markers, including cancer antigen 125 (CA125, 0.842) and human epididymis protein 4 (HE4, 0.717). Notably, our method exhibited remarkable efficacy in differentiating early-stage ovarian clear cell carcinoma from endometrioma, achieving a ROC-AUC of 0.808, outperforming CA125 (0.538) and HE4 (0.557). Our study presents a promising breakthrough in the early detection of EOC through the innovative CSGSA method. The integration of glycan alterations with cancer-related tumor markers has demonstrated exceptional diagnostic potential.

Abstract Deep learning techniques were used in ophthalmology to develop artificial intelligence (AI) models for predicting the short-term effectiveness of anti-VEGF therapy in patients with macular edema secondary to branch retinal vein occlusion (BRVO-ME). 180 BRVO-ME patients underwent pre-treatment FFA scans. After 3 months of ranibizumab injections, CMT measurements were taken at baseline and 1-month intervals. Patients were categorized into good and poor prognosis groups based on macular edema at the 4th month follow-up. FFA-Net, a VGG-based classification network, was trained using FFA images from both groups. Class activation heat maps highlighted important locations. Benchmark models (DesNet-201, MobileNet-V3, ResNet-152, MansNet-75) were compared for training results. Performance metrics included accuracy, sensitivity, specificity, F1 score, and ROC curves. FFA-Net predicted BRVO-ME treatment effect with an accuracy of 88.63% and an F1 score of 0.89, with a sensitivity and specificity of 79.40% and 71.34%, respectively.The AUC of the ROC curve for the FFA-Net model was 0.71. The use of FFA based on deep learning technology has feasibility in predicting the treatment effect of BRVO-ME. The FFA-Net model constructed with the VGG model as the main body has good results in predicting the treatment effect of BRVO-ME. The typing of BRVO in FFA may be an important factor affecting the prognosis.

Abstract Autonomous Vehicles (AV’s) have achieved more popularity in vehicular technology in recent years. For the development of secure and safe driving, these AV’s help to reduce the uncertainties such as crashes, heavy traffic, pedestrian behaviours, random objects, lane detection, different types of roads and their surrounding environments. In AV’s, Lane Detection is one of the most important aspects which helps in lane holding guidance and lane departure warning. From Literature, it is observed that existing deep learning models perform better on well maintained roads and in favourable weather conditions. However, performance in extreme weather conditions and curvy roads need focus. The proposed work focuses on presenting an accurate lane detection approach on poor roads, particularly those with curves, broken lanes, or no lane markings and extreme weather conditions. Lane Detection with Convolutional Attention Mechanism (LD-CAM) model is proposed to achieve this outcome. The proposed method comprises an encoder, an enhanced convolution block attention module (E-CBAM), and a decoder. The encoder unit extracts the input image features, while the E-CBAM focuses on quality of feature maps in input images extracted from the encoder, and the decoder provides output without loss of any information in the original image. The work is carried out using the distinct data from three datasets called Tusimple for different weather condition images, Curve Lanes for different curve lanes images and Cracks and Potholes for damaged road images. The proposed model trained using these datasets showcased an improved performance attaining an Accuracy of 97.90%, Precision of 98.92%, F1-Score of 97.90%, IoU of 98.50% and Dice Co-efficient as 98.80% on both structured and defective roads in extreme weather conditions.

Abstract Epileptic seizure (ES) detection is an active research area, that aims at patient-specific ES detection with high accuracy from electroencephalogram (EEG) signals. The early detection of seizure is crucial for timely medical intervention and prevention of further injuries of the patients. This work proposes a robust deep learning framework called HyEpiSeiD that extracts self-trained features from the pre-processed EEG signals using a hybrid combination of convolutional neural network followed by two gated recurrent unit layers and performs prediction based on those extracted features. The proposed HyEpiSeiD framework is evaluated on two public datasets, the UCI Epilepsy and Mendeley datasets. The proposed HyEpiSeiD model achieved 99.01% and 97.50% classification accuracy, respectively, outperforming most of the state-of-the-art methods in epilepsy detection domain.

In e-commerce platforms, sentiment analysis on an enormous number of user reviews efficiently enhances user satisfaction. In this article, an automated product recommendation system is developed based on machine and deep-learning models. In the initial step, the text data are acquired from the Amazon Product Reviews dataset, which includes 60,000 customer reviews with 14,806 neutral reviews, 19,567 negative reviews, and 25,627 positive reviews. Further, the text data denoising is carried out using techniques such as stop word removal, stemming, segregation, lemmatization, and tokenization. Removing stop-words (duplicate and inconsistent text) and other denoising techniques improves the classification performance and decreases the training time of the model. Next, vectorization is accomplished utilizing the term frequency-inverse document frequency technique, which converts denoised text to numerical vectors for faster code execution. The obtained feature vectors are given to the modified convolutional neural network model for sentiment analysis on e-commerce platforms. The empirical result shows that the proposed model obtained a mean accuracy of 97.40% on the APR dataset.

Abstract This study aimed to (1) replicate a deep-learning-based model for cerebral aneurysm segmentation in TOF-MRAs, (2) improve the approach by testing various fully automatic pre-processing pipelines, and (3) rigorously validate the model’s transferability on independent, external test-datasets. A convolutional neural network was trained on 235 TOF-MRAs acquired on local scanners from a single vendor to segment intracranial aneurysms. Different pre-processing pipelines including bias field correction, resampling, cropping and intensity-normalization were compared regarding their effect on model performance. The models were tested on independent, external same-vendor and other-vendor test-datasets, each comprised of 70 TOF-MRAs, including patients with and without aneurysms. The best-performing model achieved excellent results on the external same-vendor test-dataset, surpassing the results of the previous publication with an improved sensitivity (0.97 vs. ~ 0.86), a higher Dice score coefficient (DSC, 0.60 ± 0.25 vs. 0.53 ± 0.31), and an improved false-positive rate (0.87 ± 1.35 vs. ~ 2.7 FPs/case). The model further showed excellent performance in the external other-vendor test-datasets (DSC 0.65 ± 0.26; sensitivity 0.92, 0.96 ± 2.38 FPs/case). Specificity was 0.38 and 0.53, respectively. Raising the voxel-size from 0.5 × 0.5×0.5 mm to 1 × 1×1 mm reduced the false-positive rate seven-fold. This study successfully replicated core principles of a previous approach for detecting and segmenting cerebral aneurysms in TOF-MRAs with a robust, fully automatable pre-processing pipeline. The model demonstrated robust transferability on two independent external datasets using TOF-MRAs from the same scanner vendor as the training dataset and from other vendors. These findings are very encouraging regarding the clinical application of such an approach.

Abstract This study investigated the usefulness of deep learning-based automatic detection of temporomandibular joint (TMJ) effusion using magnetic resonance imaging (MRI) in patients with temporomandibular disorder and whether the diagnostic accuracy of the model improved when patients’ clinical information was provided in addition to MRI images. The sagittal MR images of 2948 TMJs were collected from 1017 women and 457 men (mean age 37.19 ± 18.64 years). The TMJ effusion diagnostic performances of three convolutional neural networks (scratch, fine-tuning, and freeze schemes) were compared with those of human experts based on areas under the curve (AUCs) and diagnosis accuracies. The fine-tuning model with proton density (PD) images showed acceptable prediction performance (AUC = 0.7895), and the from-scratch (0.6193) and freeze (0.6149) models showed lower performances (p

Abstract This study aimed to investigate the performance and factors affecting the species classification of convolutional neural network (CNN) architecture using whole-part and earlywood-part cross-sectional datasets of six Korean Quercus species. The accuracy of species classification for each condition was analyzed using the datasets, data augmentation, and optimizers—stochastic gradient descent (SGD), adaptive moment estimation (Adam), and root mean square propagation (RMSProp)—based on a CNN architecture with three to four convolutional layers. The model trained with the augmented dataset yielded significantly superior results in terms of classification accuracy compared to the model trained with the non-augmented dataset. The augmented dataset was the only factor affecting classification accuracy in the final five epochs. In contrast, four factors in the entire epoch, such as the Adam and SGD optimizers and the earlywood-part and whole-part datasets, affected species classification. The arrangement of earlywood vessels, broad rays, and axial parenchyma was identified as a major influential factor in the CNN species classification using gradient-weighted class activation mapping (Grad-CAM) analysis. The augmented whole-part dataset with the Adam optimizer achieved the highest classification accuracy of 85.7% during the final five epochs of the test phase.

Person identification through chest X-ray radiographs stands as a vanguard in both healthcare and biometrical security domains. In contrast to traditional biometric modalities, such as facial recognition, fingerprints and iris scans, the research orientation towards chest X-ray recognition has been spurred by its remarkable recognition rates. Capturing the intricate anatomical nuances of an individual's rib cage, lungs and heart, chest X-ray images emerge as a focal point for identification, even in scenarios where the human body is entirely damaged. Concerning the field of deep learning, a paradigm is exemplified in contemporary generations, with promising outcomes in classification and image similarity challenges. However, the training of convolutional neural networks (CNNs) requires copious labelled data and is time-consuming. In this study, we delve into the rich repository of the NIH ChestX-ray14 dataset, comprising 112,120 frontal-view chest radiographs from 30,805 unique patients. Our methodology is nuanced, employing the potency of Siamese neural networks and the triplet loss in conjunction with refined CNN models for feature extraction. The Siamese networks facilitate robust image similarity comparison, while the triplet loss optimizes the embedding space, mitigating intra-class variations and amplifying inter-class distances. A meticulous examination of our experimental results reveals profound insights into our model performance. Noteworthy is the remarkable accuracy achieved by the VGG-19 model, standing at an impressive 97%. This achievement is underpinned by a well-balanced precision of 95.3% and an outstanding recall of 98.4%. Surpassing other CNN models utilized in our research and outshining existing state-of-the-art models, our approach establishes itself as a vanguard in the pursuit of person identification through chest X-ray images.

Abstract Autism Spectrum Disorder (ASD) is a chronic condition characterised by impairments in social interaction and communication. Early detection of ASD is desired, and there exists a demand for the development of diagnostic aids to facilitate this. A lightweight Involutional Neural Network (INN) architecture has been developed to diagnose ASD. The model follows a simpler architectural design and has less number of parameters than the state‐of‐the‐art (SOTA) image classification models, requiring lower computational resources. The proposed model is trained to detect ASD from eye‐tracking scanpath (SP), heatmap (HM), and fixation map (FM) images. Monte Carlo Dropout has been applied to the model to perform an uncertainty analysis and ensure the effectiveness of the output provided by the proposed INN model. The model has been trained and evaluated using two publicly accessible datasets. From the experiment, it is seen that the model has achieved 98.12% accuracy, 96.83% accuracy, and 97.61% accuracy on SP, FM, and HM, respectively, which outperforms the current SOTA image classification models and other existing works conducted on this topic.

Abstract Purpose In pediatric medicine, precise estimation of bone age is essential for skeletal maturity evaluation, growth disorder diagnosis, and therapeutic intervention planning. Conventional techniques for determining bone age depend on radiologists’ subjective judgments, which may lead to non-negligible differences in the estimated bone age. This study proposes a deep learning-based model utilizing a fully connected convolutional neural network(CNN) to predict bone age from left-hand radiographs. Methods The data set used in this study, consisting of 473 patients, was retrospectively retrieved from the PACS (Picture Achieving and Communication System) of a single institution. We developed a fully connected CNN consisting of four convolutional blocks, three fully connected layers, and a single neuron as output. The model was trained and validated on 80% of the data using the mean-squared error as a cost function to minimize the difference between the predicted and reference bone age values through the Adam optimization algorithm. Data augmentation was applied to the training and validation sets yielded in doubling the data samples. The performance of the trained model was evaluated on a test data set (20%) using various metrics including, the mean absolute error (MAE), median absolute error (MedAE), root-mean-squared error (RMSE), and mean absolute percentage error (MAPE). The code of the developed model for predicting the bone age in this study is available publicly on GitHub at https://github.com/afiosman/deep-learning-based-bone-age-estimation . Results Experimental results demonstrate the sound capabilities of our model in predicting the bone age on the left-hand radiographs as in the majority of the cases, the predicted bone ages and reference bone ages are nearly close to each other with a calculated MAE of 2.3 [1.9, 2.7; 0.95 confidence level] years, MedAE of 2.1 years, RMAE of 3.0 [1.5, 4.5; 0.95 confidence level] years, and MAPE of 0.29 (29%) on the test data set. Conclusion These findings highlight the usability of estimating the bone age from left-hand radiographs, helping radiologists to verify their own results considering the margin of error on the model. The performance of our proposed model could be improved with additional refining and validation.

Introduction: Ultrasonography (US) features of papillary thyroid cancers (PTCs) are used to select nodules for biopsy due to their association with tumor behavior. However, the molecular biological mechanisms that lead to the characteristic US features of PTCs are largely unknown. Objectives: This study aimed to investigate the molecular biological mechanisms behind US features assessed by radiologists and three convolutional neural networks (CNN) through transcriptome analysis. Methods: Transcriptome data from 273 PTC tissue samples were generated and differentially expressed genes (DEGs) were identified according to US feature. Pathway enrichment analyses were also conducted by gene set enrichment analysis (GSEA) and ClusterProfiler according to assessments made by radiologists and three CNNs - CNN1 (ResNet50), CNN2 (ResNet101) and CNN3 (VGG16). Signature gene scores for PTCs were calculated by single-sample GSEA (ssGSEA). Results: Individual suspicious US features consistently suggested an upregulation of genes related to immune response and epithelial-mesenchymal transition (EMT). Likewise, PTCs assessed as positive by radiologists and three CNNs showed the coordinate enrichment of similar gene sets with abundant immune and stromal components. However, PTCs assessed as positive by radiologists had the highest number of DEGs, and those assessed as positive by CNN3 had more diverse DEGs and gene sets compared to CNN1 or CNN2. The percentage of PTCs assessed as positive or negative concordantly by radiologists and three CNNs was 85.6% (231/273) and 7.1% (3/273), respectively. Conclusion: US features assessed by radiologists and CNNs revealed molecular biologic features and tumor microenvironment in PTCs.

The dicentric chromosome assay is a gold standard method to estimate radiation exposure by calculating the ratio of dicentric chromosomes existing in cells. The objective of this study was to propose an automatic dicentric chromosome discrimination method based on deep convolutional neural networks using radiation exposure patient data. From 45 patients with radiation exposure, conventional Giemsa-stained images of 116,258 normal and 2800 dicentric chromosomes were confirmed. ImageNet was used to pre-train VGG19, which was modified and fine-tuned. The proposed modified VGG19 demonstrated dicentric chromosome discrimination performance, with a true positive rate of 0.927, a true negative rate of 0.997, a positive predictive value of 0.882, a negative predictive value of 0.998, and an area under the receiver operating characteristic curve of 0.997.

During the decommissioning of nuclear and particle accelerator facilities, a considerable amount of large-scale radioactive waste may be generated. Accurately defining the activation level of the waste is crucial for proper disposal. However, directly measuring the internal radioactivity distribution poses challenges. This study introduced a novel technology employing machine learning to assess the internal radioactivity distribution based on external measurements. Random radioactivity distribution within a structure were established, and the photon spectrum measured by detectors from outside the structure was simulated using the FLUKA Monte-Carlo code. Through training with spectrum data corresponding to various radioactivity distributions, an evaluation model for radioactivity using simulated data was developed by above Monte-Carlo simulation. Convolutional Neural Network and Transformer methods were utilized to establish the evaluation model. The machine learning construction involves 5425 simulation datasets, and 603 datasets, which were used to obtain the evaluated results. Preprocessing was applied to the datasets, but the evaluation model using raw spectrum data showed the best evaluation results. The estimation of the intensity and shape of the radioactivity distribution inside the structure was achieved with a relative error of 10%. Additionally, the evaluation based on the constructed model takes only a few seconds to complete the process.

In the domain of clinical imaging, the exact and quick identification proof of bone fractures assumes a crucial part in a pivotal role in facilitating timely and effective patient care. This research tends to this basic need by harnessing the force of profound learning, explicitly utilizing a Convolutional Neural Network (CNN) model as the foundation of our technique. The essential target of our study was to improve the mechanized recognition of bone fractures in X-ray images, utilizing the capacities of deep learning algorithms. The use of a CNN model permitted us to successfully capture and learn intricate patterns and features within the X-ray images, empowering the framework to make exact fracture detections. The training process included presenting the model to a various dataset, guaranteeing its versatility to an extensive variety of fracture types. The results of our research show the excellent performance of the CNN model in fracture detection, where our model has achieved an Average Precision 89.5%, Average Recall 87%, and the overall Accuracy 91%. These metrics assert the vigour of our methodology and highlight the capability of deep learning in medical image analysis.

The agriculture industry has an enormous influence on a nation's economy. Loss of yield due to plant diseases remains a reason, reducing crop quantity and quality. Incorrect diagnosis of crop diseases can result in improper application of chemical pesticides, which promotes immune microbial strains, raises expenses, and triggers fresh outbreaks that are harmful to the economy and the ecosystem. Despite the potential of Machine Learning (ML) and Deep Learning (DL) approaches in plant disease detection, their limited effectiveness results in poor or late disease detection. Resolving this issue is critical, requiring the development of more accurate disease detection methods. This research introduces an innovative approach for the detection of apple leaf diseases utilizing the CNN-based Inception-v3 model. The dataset comprises images taken on location without having any control over the image-capturing settings may provide better relevance to real-world scenarios. The proposed method integrates canny edge detection and watershed transformation to achieve accurate image segmentation, thereby enhancing the identification of disease regions. Additionally, exploratory data analysis was performed, and channel distributions were visualized to understand the dataset's characteristics. To ensure robust evaluation, the model's performance underwent stratified 5-fold cross-validation. The model classified plant images with 84.60% precision, 87.40% recall, 85.00% F1-score, and 94.76% accuracy. Experimental results substantiate the efficacy of the proposed approach, surpassing existing methods in disease classification.

The accurate prediction of friction coefficients is crucial for the maintenance of sliding mechanical components to enable the timely detection of potential failures. Traditional methods rely on sensors like load cells and strain gauges to measure friction coefficients. However, these conventional techniques face challenges in real-time measurement during machine operation owing to physical constraints associated with sensor placement. To address this limitation, this study investigates the application of laser speckle patterns for predicting friction coefficients through a novel approach using convolutional neural networks (CNNs). The laser speckle technique offers rich surface condition data, while CNNs, which are particularly advanced in managing vast datasets, excel in establishing relationships between diverse factors for precise inference, classification, and prediction. Utilizing ResNet, a leading CNN architecture, a new friction tester capable of concurrently recording friction coefficients and speckle patterns in a cylinder-on-disk friction test was developed. The findings reveal that the CNN-based method, especially with ResNet, attained a coefficient of determination (R2) of 0.758, demonstrating its effectiveness in the accurate prediction of friction coefficients. This study significantly advances the field of friction coefficient prediction, highlighting the innovative application of laser speckle methodologies combined with machine learning techniques.

ObjectiveThere exist a high genetic correlation among various morphological characteristics of Mongolian horses. Utilizing advanced technology to obtain body structure parameters related to athletic performance could provide data support for breeding institutions to develop scientific breeding plans and establish the groundwork for further improvement of Mongolian horse breeds. However, traditional manual measurement methods are time-consuming, labor-intensive, and may cause certain stress responses in horses. Therefore, ensuring precise and effective measurement of Mongolian horse body dimensions is crucial for formulating early breeding plans.MethodVideo images of 50 adult Mongolian horses in the suitable breeding stage at the Inner Mongolia Agricultural University Horse Breeding Technical Center was first collected. Fifty images per horse were captured to construct the training and validation sets, resulting in a total of 2 500 high-definition RGB images of Mongolian horses, with an equal ratio of images depicting horses in motion and at rest. To ensure the model's robustness and considering issues such as angles, lighting, and image blurring during actual image capture, a series of enhancement algorithms were applied to the original dataset, expanding the Mongolian horse image dataset to 4 000 images. The YOLOv8n-pose was employed as the foundational keypoint detection model. Through the design of the C2f_DCN module, deformable convolution (DCNV2) was integrated into the C2f module of the Backbone network to enhance the model's adaptability to different horse poses in real-world scenes. Besides, an SA attention module was added to the Neck network to improve the model's focus on critical features. The original loss function was replaced with SCYLLA-IoU (SIoU) to prioritize major image regions, and a cosine annealing method was employed to dynamically adjust the learning rate during model training. The improved model was named DSS-YOLO (DCNv2-SA-SIoU-YOLO) network model. Additionally, a test set comprising 30 RGB-D images of mature Mongolian horses was selected for constructing body dimension measurement tasks. DSS-YOLO was used for keypoint detection of body dimensions. The 2D keypoint coordinates from RGB images were fused with corresponding depth values from depth images to obtain 3D keypoint coordinates, and Mongolian horse's point cloud information was transformed. Point cloud processing and analysis were performed using pass-through filtering, random sample consensus (RANSAC) shape fitting, statistical outlier filtering, and principal component analysis (PCA) coordinate system correction. Finally, body height, body oblique length, croup height, chest circumference, and croup circumference were automatically computed based on keypoint spatial coordinates.Results and DiscussionThe proposed DSS-YOLO model exhibited parameter and computational costs of 3.48 M and 9.1 G, respectively, with an average accuracy mAP0.5:0.95 reaching 92.5%, and a dDSS of 7.2 pixels. Compared to Hourglass, HRNet, and SimCC, mAP0.5:0.95 increased by 3.6%, 2.8%, and 1.6%, respectively. By relying on keypoint coordinates for automatic calculation of body dimensions and suggesting the use of a mobile least squares curve fitting method to complete the horse's hip point cloud, experiments involving 30 Mongolian horses showed a mean average error (MAE) of 3.77 cm and mean relative error (MRE) of 2.29% in automatic measurements.ConclusionsThe results of this study showed that DSS-YOLO model combined with three-dimensional point cloud processing methods can achieve automatic measurement of Mongolian horse body dimensions with high accuracy. The proposed measurement method can also be extended to different breeds of horses, providing technical support for horse breeding plans and possessing practical application value.

In order to investigate the quality changes and shelf life of blueberries stored in different temperature, 21 quality indexes, including color parameters, mass loss rate, spoilage rate and texture parameters, were measured on “Freedom” blueberries at three storage temperatures (0, 4 and 25 ℃). Using five machine learning algorithms with a self-contained function of feature selection, seven key features affecting the shelf life were selected as input variables to construct a shelf life prediction model using gated recurrent unit (GRU) alone or in combination with convolutional neural network (CNN) and/or attention (AE) mechanism. The results showed that compared with the GRU model, the mean absolute error (MAE), mean square error (MSE) and mean absolute percentage error (MAPE) of the CNN-GRU-AE model decreased by 75.83%, 91.46%, 61.58%, respectively, and the coefficient and determination increased by 2.25%, indicating significantly improved accuracy of shelf-life prediction. This study provides theoretical support for the shelf life prediction of blueberries at different storage temperatures.

This research classifies acoustic instruments using Convolutional Neural Network (CNN). We utilize a dataset from Kaggle containing audio recordings of piano, violin, drums, and guitar. The training set consists of 700 guitar, percussion, violin, and 528 piano samples. The test set contains 80 samples of each instrument. Features such as Mel spectrograms, MFCCs, and other spectral and non-spectral characteristics are extracted using the Librosa package. Three feature sets—spectral-only, non-spectral-only, and a combined set—are employed to evaluate the efficacy of CNN models. Various CNN configurations are tested by adjusting the number of convolutional filters, learning rates, and epochs. The combined feature set achieves the highest performance, with a validation accuracy of 71.8% and a training accuracy of 76.9%. In comparison, non-spectral features achieve a validation accuracy of 68.4%, and spectral-only features achieve 69.3%. These findings highlight the benefits of using a comprehensive feature set for accurate classification.

Tires are a very important component in a vehicle because they are related to driving safety. Defective tires often cause accidents ranging from minor to fatal accidents. Convolutional Neural Network (CNN) is a type of neural network that is used to detect and recognize objects in an image. CNN can imitate the image recognition system in the human visual cortex, making it suitable for identification and classification of image data. This research aims to develop and evaluate a CNN model that is able to classify vehicle tires as 'defective' or 'good'. Model uses a total of 1856 tire images from kaggle.com and is labeled 'defective' or 'good'. Dataset is split using four different data split ratios (60:40, 70:30, 80:20, and 90:10) to determine the optimal distribution that improves the generalization ability of the model. Model evaluation uses accuracy, precision and recall matrices, which are calculated based on the confusion matrix results from testing on 300 data samples. Research results show that the model achieves the best performance at a split ratio of 80:20, with an accuracy of 76.67%, precision of 77.33%, and recall of 76.32%.

Abstract Purpose objective(s) Manual contouring of the prostate region in planning computed tomography (CT) images is a challenging task due to factors such as low contrast in soft tissues, inter- and intra-observer variability, and variations in organ size and shape. Consequently, the use of automated contouring methods can offer significant advantages. In this study, we aimed to investigate automated male pelvic multi-organ contouring in multi-center planning CT images using a hybrid convolutional neural network-vision transformer (CNN-ViT) that combines convolutional and ViT techniques. Materials/methods We used retrospective data from 104 localized prostate cancer patients, with delineations of the clinical target volume (CTV) and critical organs at risk (OAR) for external beam radiotherapy. We introduced a novel attention-based fusion module that merges detailed features extracted through convolution with the global features obtained through the ViT. Results The average dice similarity coefficients (DSCs) achieved by VGG16-UNet-ViT for the prostate, bladder, rectum, right femoral head (RFH), and left femoral head (LFH) were 91.75%, 95.32%, 87.00%, 96.30%, and 96.34%, respectively. Experiments conducted on multi-center planning CT images indicate that combining the ViT structure with the CNN network resulted in superior performance for all organs compared to pure CNN and transformer architectures. Furthermore, the proposed method achieves more precise contours compared to state-of-the-art techniques. Conclusion Results demonstrate that integrating ViT into CNN architectures significantly improves segmentation performance. These results show promise as a reliable and efficient tool to facilitate prostate radiotherapy treatment planning.

Abstract Virus‒host protein‒lncRNA interaction (VHPLI) predictions are critical for decoding the molecular mechanisms of viral pathogens and host immune processes. Although VHPLI interactions have been predicted in both plants and animals, they have not been extensively studied in viruses. For the first time, we propose a new deep learning-based approach that consists mainly of a convolutional neural network and bidirectional long and short-term memory network modules in combination with transfer learning named CBIL‒VHPLI to predict viral–host protein‒lncRNA interactions. The models were first trained on large and diverse datasets (including plants, animals, etc.). Protein sequence features were extracted using a k-mer method combined with the one-hot encoding and composition–transition–distribution (CTD) methods, and lncRNA sequence features were extracted using a k-mer method combined with the one-hot encoding and Z curve methods. The results obtained on three independent external validation datasets showed that the pre-trained CBIL‒VHPLI model performed the best with an accuracy of approximately 0.9. Pretraining was followed by conducting transfer learning on a viral protein–human lncRNA dataset, and the fine-tuning results showed that the accuracy of CBIL‒VHPLI was 0.946, which was significantly greater than that of the previous models. The final case study results showed that CBIL‒VHPLI achieved a prediction reproducibility rate of 91.6% for the RIP-Seq experimental screening results. This model was then used to predict the interactions between human lncRNA PIK3CD-AS2 and the nonstructural protein 1 (NS1) of the H5N1 virus, and RNA pull-down experiments were used to prove the prediction readiness of the model in terms of prediction. The source code of CBIL‒VHPLI and the datasets used in this work are available at https://github.com/Liu-Lab-Lnu/CBIL-VHPLI for academic usage.

Abstract The bone marrow overproduces immature cells in the malignancy known as Acute Lymphoblastic Leukemia (ALL). In the United States, about 6500 occurrences of ALL are diagnosed each year in both children and adults, comprising nearly 25% of pediatric cancer cases. Recently, many computer-assisted diagnosis (CAD) systems have been proposed to aid hematologists in reducing workload, providing correct results, and managing enormous volumes of data. Traditional CAD systems rely on hematologists’ expertise, specialized features, and subject knowledge. Utilizing early detection of ALL can aid radiologists and doctors in making medical decisions. In this study, Deep Dilated Residual Convolutional Neural Network (DDRNet) is presented for the classification of blood cell images, focusing on eosinophils, lymphocytes, monocytes, and neutrophils. To tackle challenges like vanishing gradients and enhance feature extraction, the model incorporates Deep Residual Dilated Blocks (DRDB) for faster convergence. Conventional residual blocks are strategically placed between layers to preserve original information and extract general feature maps. Global and Local Feature Enhancement Blocks (GLFEB) balance weak contributions from shallow layers for improved feature normalization. The global feature from the initial convolution layer, when combined with GLFEB-processed features, reinforces classification representations. The Tanh function introduces non-linearity. A Channel and Spatial Attention Block (CSAB) is integrated into the neural network to emphasize or minimize specific feature channels, while fully connected layers transform the data. The use of a sigmoid activation function concentrates on relevant features for multiclass lymphoblastic leukemia classification The model was analyzed with Kaggle dataset (16,249 images) categorized into four classes, with a training and testing ratio of 80:20. Experimental results showed that DRDB, GLFEB and CSAB blocks’ feature discrimination ability boosted the DDRNet model F1 score to 0.96 with minimal computational complexity and optimum classification accuracy of 99.86% and 91.98% for training and testing data. The DDRNet model stands out from existing methods due to its high testing accuracy of 91.98%, F1 score of 0.96, minimal computational complexity, and enhanced feature discrimination ability. The strategic combination of these blocks (DRDB, GLFEB, and CSAB) are designed to address specific challenges in the classification process, leading to improved discrimination of features crucial for accurate multi-class blood cell image identification. Their effective integration within the model contributes to the superior performance of DDRNet.

Abstract The uncertainty of true labels in medical images hinders diagnosis owing to the variability across professionals when applying deep learning models. We used deep learning to obtain an optimal convolutional neural network (CNN) by adequately annotating data for oral exfoliative cytology considering labels from multiple oral pathologists. Six whole-slide images were processed using QuPath for segmenting them into tiles. The images were labeled by three oral pathologists, resulting in 14,535 images with the corresponding pathologists’ annotations. Data from three pathologists who provided the same diagnosis were labeled as ground truth (GT) and used for testing. We investigated six models trained using the annotations of (1) pathologist A, (2) pathologist B, (3) pathologist C, (4) GT, (5) majority voting, and (6) a probabilistic model. We divided the test by cross-validation per slide dataset and examined the classification performance of the CNN with a ResNet50 baseline. Statistical evaluation was performed repeatedly and independently using every slide 10 times as test data. For the area under the curve, three cases showed the highest values (0.861, 0.955, and 0.991) for the probabilistic model. Regarding accuracy, two cases showed the highest values (0.988 and 0.967). For the models using the pathologists and GT annotations, many slides showed very low accuracy and large variations across tests. Hence, the classifier trained with probabilistic labels provided the optimal CNN for oral exfoliative cytology considering diagnoses from multiple pathologists. These results may lead to trusted medical artificial intelligence solutions that reflect diverse diagnoses of various professionals.

Abstract This paper proposes a geometric-based technique for compressing convolutional neural networks to accelerate computations and improve generalization by eliminating non-informative components. The technique utilizes a geometric index called separation index to evaluate the functionality of network elements such as layers and filters. By applying this index along with center-based separation index, a systematic algorithm is proposed that optimally compresses convolutional and fully connected layers. The algorithm excludes layers with low performance, selects the best subset of filters in the filtering layers, and tunes the parameters of fully connected layers using center-based separation index. An illustrative example of classifying CIFAR-10 dataset is presented to explain the algorithm step-by-step. The proposed method achieves impressive pruning results on networks trained by CIFAR-10 and ImageNet datasets, with 87.5%, 77.6%, and 78.8% of VGG16, GoogLeNet, and DenseNet parameters pruned, respectively. Comparisons with state-of-the-art works are provided to demonstrate the effectiveness of the proposed method.

Abstract Background Pepper Phytophthora blight is a devastating disease during the growth process of peppers, significantly affecting their yield and quality. Accurate, rapid, and non-destructive early detection of pepper Phytophthora blight is of great importance for pepper production management. This study investigated the possibility of using multispectral imaging combined with machine learning to detect Phytophthora blight in peppers. Peppers were divided into two groups: one group was inoculated with Phytophthora blight, and the other was left untreated as a control. Multispectral images were collected at 0-h samples before inoculation and at 48, 60, 72, and 84 h after inoculation. The supporting software of the multispectral imaging system was used to extract spectral features from 19 wavelengths, and textural features were extracted using a gray-level co-occurrence matrix (GLCM) and a local binary pattern (LBP). The principal component analysis (PCA), successive projection algorithm (SPA), and genetic algorithm (GA) were used for feature selection from the extracted spectral and textural features. Two classification models were established based on effective single spectral features and significant spectral textural fusion features: a partial least squares discriminant analysis (PLS_DA) and one-dimensional convolutional neural network (1D-CNN). A two-dimensional convolutional neural network (2D-CNN) was constructed based on five principal component (PC) coefficients extracted from the spectral data using PCA, weighted, and summed with 19-channel multispectral images to create new PC images. Results The results indicated that the models using PCA for feature selection exhibit relatively stable classification performance. The accuracy of PLS-DA and 1D-CNN based on single spectral features is 82.6% and 83.3%, respectively, at the 48h mark. In contrast, the accuracy of PLS-DA and 1D-CNN based on spectral texture fusion reached 85.9% and 91.3%, respectively, at the same 48h mark. The accuracy of the 2D-CNN based on 5 PC images is 82%. Conclusions The research indicates that Phytophthora blight infection can be detected 48 h after inoculation (36 h before visible symptoms). This study provides an effective method for the early detection of Phytophthora blight in peppers.