Search

This paper presents a new proposal of an efficient computational model of face recognition which uses cues from the distributed face recognition mechanism of the brain, and by gathering engineering equivalent of these cues from existing literature. Three distinct and widely used features – Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and Principal components (PCs) extracted from target images are used in a manner which is simple, and yet effective. The HOG and LBP features further undergo Principal Component Analysis (PCA) for dimensionality reduction. Our model uses multi-layer perceptrons (MLP) to classify these three features and fuse them to form a sparsely connected model. A computational theory is first developed by using concepts from the information processing mechanism of the brain. Extensive experiments are carried out using eight publicly available face datasets to validate our proposed model's performance in recognizing faces with extreme variation of illumination, pose angle, expression, and background. We also investigate the same mechanism because of reasons discussed later, on object recognition tasks as well. Results obtained are extremely promising when compared with other face and object recognition algorithms including CNN and deep learning-based methods. This highlights that simple computational processes, if clubbed properly, can produce competing performance with best algorithms. [ABSTRACT FROM AUTHOR]

Medical image segmentation is one of the most challenging tasks in medical image analysis which aims to extract effective information and improve the level of clinical diagnosis. In last decades, automatic segmentation based on Deep Learning (DL) models such as U-Net and CNN architectures have been widely used to automatically extract organs or lesions contours in order to overcome manual segmentation limitations. In this paper, we performed a comparative study between CNN and U-Net performance for medical image segmentation applied to cardiac-MRI segmentation using U-Net from short-axis MRI images of ACDC database. The adopted architecture was trained and tested with and without data augmentation. The obtained results show a strong agreement between the labelled masks and the predicted ones with a mean DSC that reached 97,9% and a mean Hausdorrf Distance (HD) that reached 5.318 mm. A quantitative comparison was made on two levels. The first one is an intra-comparison made between the adopted model and methods based on the same architecture and having been trained and tested using the same database which proved that our method reached the highest performance and can be considered as a promising tool for medical image segmentation. The second one is an inter-comparison made between U-Net and CNN performance which proved that U-Net is more suitable for carrying out this task since it takes less time for training as it does not have a fully connected layer and offers a fairly significant similarity to ground truth compared to CNN. [ABSTRACT FROM AUTHOR]

In the last two decades, improvement in artificial intelligence and medical imaging technology have made healthcare sector to achieve some remarkable achievements in diseases analysis and prediction. Due to advancement in medical imaging technology the brain images are taken in different modalities, that gives 3D view of different sections of brain for tumor diagnosis. The ability to extract relevant characteristics from magnetic resonance imaging (MRI) scans is a crucial step for brain tumor classifiers. As a result, several studies have proposed various strategies to extract relevant features from different modalities of MRI to predict the growth of abnormal tumor. Most of techniques used conventional techniques of image processing for feature extraction and machine learning for classification. More recently, the use deep learning algorithms in medical imaging has resulted in significant improvements in the classification and diagnosis of brain tumor. Since tumors are located at different regions of brain, the localizing the tumor and classifying it to particular category is challenging task. In this paper, we have solved this problem by designing deep ensemble model. In the proposed approach, first shallow convolutional neural network (SCNN) and VGG16 network were designed with T1C modality MRI image and subsequently loss and accuracy were examined. To improve the performance of model in terms accuracy and loss information, the extracted features from both the deep learning model were fused to improve the classification accuracy of three types of tumors. The obtained results from ensemble deep convolutional neural network model (EDCNN), proved that the fusion of deep learning model improves the accuracy of multiclass classification problem and also tries to address the problem of overfitting of model for imbalance dataset. The proposed model tries to give classification accuracy up to 97.77%. Furthermore, the proposed framework, achieves competitive results when compared with other state of art studies. [ABSTRACT FROM AUTHOR]

Now a days, Video Summarization is highly explored research area by the research community, which has off late gained a lot of interest from researchers globally. It comprises extraction of important frames from the big size video input in order to extract the important events from the video into a much smaller yet comprehensively summarized form. Towards effective video summarization, a novel methodology is proposed in this paper, that works on AoA (attention over attention) strategy. The proposed deep model is based on multiple attention modules including spatial, channel, and multi-headed attention. The AoA enables us to capture inter spatial and inter channel relationship between the features effectively. The proposed attention module is applied over the set of existing features from video summarization datasets. Progressively applying attention ensures to highlight the most important contents from the input frames, thereby producing more effective key frames. Several ablation studies have also been performed to analyze the position spatial and channel attention to determine the best possible architecture based on experiments alongside the theoretical proofs of AoA architecture. For the experimental work, two benchmark dataset have been used and compared the performance with existing methods. [ABSTRACT FROM AUTHOR]

Agricultural sector is one of the main sectors in India. Productive growth and high yield production of fruits is important and required for the agricultural industry. Application of computer vision and image processing has aid agriculture to boost yield rating, disease spotting, irrigation, fruit classification and maturity grading. Computer vision and image processing approach can be used to lower the time utilization and has made it economical. A convolutional neural network (CNN) is a type of artificial neural network used in image identification and processing that is mainly planned to process pixel data. This paper utilized a convolutional neural network (CNN) to classify bananas into raw or ripe without requiring labour. Bananas were classified by their colour appearance in the CNN classification and not only this it can be also used to classify many other types of fruits and vegetables. [ABSTRACT FROM AUTHOR]

Automation in the agricultural field is a requirement of all the countries. Usually, plant diseases are observed in the form of visual symptoms and many deep learning-based models have achieved outstanding results in the classification of plant leaf diseases in recent years. Beans plant diseases like bean rust disease and angular leaf spot disease reduce the bean crop yield. To treat the problem at an early stage, an appropriate diagnosis for this crop disease is required. In this paper, three deep learning-based pre-trained models namely MobileNetV2, EfficientNetB6, and NasNet were used to perform transfer learning on the Beans Leaf image dataset containing 1295 images with three different classes. Furthermore, different optimization techniques were also used to highlight the variation in performance of different Convolutional Neural Network (CNN) models. The analysis of experimental results shows that EfficientNetB6 performs better with 91.74% accuracy than other models. This study would be helpful to understand the role of different optimizers on the CNN models. Furthermore, agricultural scientists could employ a real-time-based application of the best-suited model for the farmers to adopt prevention measures in disease-vulnerable areas. As a result, prompt action would aid in minimizing plant productivity loss. It will further revenue the economic growth and agricultural productivity. [ABSTRACT FROM AUTHOR]

The notion of smart farming is gaining traction in the agricultural industry these days, and it makes use of sensors and a variety of machine learning based technologies. According to recent surveys, 56 percent of the agricultural industry is facing significant losses because of diseases developing on plant leaves. It's critical to keep track of the disease's spread and enhance agricultural yields. To prevent the disease from spreading, we must first recognize it on time and prevent it. As a result, we may solve this problem by putting in place some algorithms for detecting sickness on leaves. This paper presents a comparative analysis between support vector machines (SVM) model, K-Nearest Neighbor (KNN) model and convolution neural network (CNN) model. The three different models are presented and examined in this research, and they can detect eight different leaf diseases. The CNN model has achieved an accuracy of 96 percent when trained with the images of soyabean leaf disease dataset, outperforms the KNN and SVM models, which have accuracy of 64 percent and 76 percent, respectively [ABSTRACT FROM AUTHOR]

Human-Computer Interaction (HCI) in an intelligent way, which aims at creating scalable and flexible solutions. Big tech firms and businesses believe in the success of HCI as it allows them to profit from on-demand technology and infrastructure for information-centric applications without having to use public clouds. Because of its capacity to imitate human coding abilities, facial expression recognition and software-based facial expression identification systems are crucial. This paper proposes a system of recognizing the emotional condition of humans, given a facial expression, and conveys two methods of predicting the age and gender factors from human faces. This research also aims in understanding the influences posed by gender and age of humans on their facial expressions. The model can currently detect 7 emotions based on the facial data of a person - (Anger, Disgust, Happy, Fear, Sad, Surprise, and Neutral state). The proposed system is divided into three segments: a.) Gender Detection b.) Age Detection c.) Emotion Recognition. The initial model is created using 2 algorithms - KNN, and SVM. We have also utilized the architectures of some of the deep learning models such as CNN and VGG - 16 pre-trained models (Transfer Learning). The evaluation metrics show the model performance regarding the accuracy of the Recognition system. Future enhancements of this work can include the deployment of the DL and ML model onto an android or a wearable device such as a smartphone or a watch for a real-time use case. [ABSTRACT FROM AUTHOR]

Nowadays, there has been much attention on computer vision regarding human-computer interaction, especially facial expression recognition (FER). Many researchers have explored and suggested systems for this field. In this paper, we propose the Deep Learning architecture to improve the performance of models from the previous work. Additionally, we propose the BiLSTM-CNN model, which combines our proposed CNN and BiLSTM model. Besides that, we also compare the model to our CNN and LSTM-CNN models. We conduct the experiments on the CK+ dataset and evaluate the accuracy rate of the built models. Data augmentation is used in the dataset to improve the model's performance and prevent overfitting. The results demonstrate that the BiLSTM-CNN method achieves a state-of-the-art accuracy rate compared to other methods from previous work. The highest accuracy of 99.43% is reached by the BiLSTM-CNN model with data augmentation. [ABSTRACT FROM AUTHOR]

• Now a days the major oral health issues of man is crucial an area of research. • This paper traces classification, segmentation and pre-processing of dental images. • Histogram based adaptive approach is used in the radiographic X-ray 2d dental Image. • This is useful to distinguish the foreground teeth & the regions of background bones. In the present scenario the major oral health issues of man is crucial an area of research. The Data mining techniques, image processing, and Computational intelligence techniques are playing a vital role in biomedical research. Dental image processing helps to improve the early detection and classification of the diagnostic process to make accurate decisions. The radiographic 2d dental image is widely utilized for analytic thinking of several dental disorders. In this paper traces the complete steps such as classification and segmentation as well as pre-processing of dental images has been carried out. In the pre-processing, histogram based on adaptive approach is used to stretch the contrast and equalize the brightness throughout the radiographic X-ray 2d dental Image. This operation is useful to distinguish the foreground teeth and the regions of background bones. Separation of dental 2d images into regions corresponding to the objects is a fundamental step of segmentation. The hybrid graph cut segmentation is used to segment the oral cavity and its tissues. In this research deep learning based convolution neural network (CNN) has been used to process the dental image and shows promising outcomes with 97.07% accuracy. [ABSTRACT FROM AUTHOR]

Accurately predicting wind and photovoltaic power is one of the keys to improving the economy of wind-solar complementary power generation system, reducing scheduling costs and no-load losses, and ensuring grid stability. However, the natural properties of energy result in complex fluctuations in their corresponding power sequences, making accurate predictions difficult. Therefore, this paper proposes an intelligent prediction system that combines decomposition algorithms and deep learning for ultra-short-term prediction of wind and photovoltaic power. First, an improved decomposition algorithm is proposed, based on fuzzy entropy's property that its value increases with the increase of sequence uncertainty, particle swarm optimization (PSO) is used to search for the optimal parameter combinations of variational modal decomposition (VMD), so that it can automatically adjust the parameters for energy data with different characteristics to reduce the human error. Then, a convolutional neural network (CNN) architecture that balances operational efficiency and prediction performance is constructed, and the hyperparameters of the CNN are optimized using the wild horse optimization algorithm (WHO) to improve the stability and accuracy of the prediction model. In this paper, real data from wind power plants and photovoltaic power plants in China are used as experimental objects, and experiments are carried out in three aspects, namely, benchmark model selection, decomposition algorithm comparison and combined model comparison. The results show that selecting CNN as the benchmark model is a good choice; the improved VMD has better decomposition performance than other state-of-the-art decomposition algorithms. The system proposed in this paper is highly generalizable and adaptive, and its prediction performance and accuracy greatly outperform that of other comparative models, with prediction accuracies improved by 72% and 79%, respectively, compared to a single CNN model. • The hyperparameters of VMD are determined by using PSO based on fuzzy entropy. • Optimize convolutional neural network using the wild horse optimization algorithm. • The intelligent prediction system can accurately predict wind power and PV power. • Experiments based on power data from actual wind farms and PV plants. • A deep learning prediction method applied to wind and solar complementary systems. [ABSTRACT FROM AUTHOR]

• We design an EEG experimental paradigm for the diagnosis and grading of patients with cervical spondylotic myelopathy, which is a novel clinical diagnostic measurement method. • This paper proposes a deep learning network based on CNN-LSTM for classifying patients and healthy individuals. The network intelligently fuses EEG features of various frequency bands using CNN. Compared to traditional EEG decoding models, it achieves better classification results. • This paper introduces a CNN-BiLSTM-attention network that integrates the attention mechanism into CNN for grading CSM conditions. The network achieves superior performance compared to traditional EEG decoding networks. Cervical spondylotic myelopathy (CSM) is the most severe type of cervical spondylosis. It is challenging to achieve early diagnosis with current clinical diagnostic tools. In this paper, we propose an end-to-end deep learning approach for early diagnosis of CSM. Electroencephalography (EEG) experiments were conducted with patients having spinal cord cervical spondylosis and age-matched normal subjects. A Convolutional Neural Network with Long Short-Term Memory Networks (CNN-LSTM) model was employed for the classification of patients versus normal individuals. In contrast, a Convolutional Neural Network with Bidirectional Long Short-Term Memory Networks and attention mechanism (CNN-BiLSTM-attention) model was used to classify regular, mild, and severe patients. The models were trained using focal Loss instead of traditional cross-entropy Loss, and cross-validation was performed. Our method achieved a classification accuracy of 92.5 % for the two-class classification among 40 subjects and 72.2 % for the three-class classification among 36 subjects. Furthermore, we observed that the proposed model outperformed traditional EEG decoding models. This paper presents an effective computer-aided diagnosis method that eliminates the need for manual extraction of EEG features and holds potential for future auxiliary diagnosis of spinal cord-type cervical spondylosis. [ABSTRACT FROM AUTHOR]

There are lots of research papers for ASC, and in recent years it is rapidly increasing. DCASE also provides different types of competition for the submission of several papers to solve the various tasks of ASC and it is the opportunity for the research scholars either to participate in those competitions or to provide the enhanced model for ASC. This paper provides details about the various recent approaches along with the block diagram used for pre-processing required before model development for ASC. It also includes a description of different recent techniques used for the classification of different sounds for ASC tasks. The comparative analysis for different recent available techniques both for pre-processing and classification has been done and summarized in this paper. It also describes the contributions towards the survey on ASC by comparing this paper with some existing survey papers based on several parameters like functionality described separately, results from description with quantifiable value, a dataset with proper quantifiable analysis, and pictorial representation of model discussed, etc. Finally, considering the benefits for eminent research scholars, this paper has also focused on the details for future directions for both pre-processing and classification for ASC. [Display omitted] • A systematic review on pre-processing and classification techniques for acoustic scenes. • Comparative analysis of pre-processing and classification techniques. • Discussion on different processing techniques for classification. • Identifies the future directions for further research in the area. [ABSTRACT FROM AUTHOR]

Deep learning (DL) is a robust data-analysis and image-processing technique that has shown great promise in the agricultural sector. In this study, 129 papers that are based on DL applications used in agriculture are discussed, categorizing them into five areas: crop yield prediction, plant stress detection, weed and pest detection, disease detection, and smart farming. Smart farming is sub-categorized as water management, seed analysis, and soil analysis. This study highlights the potential of deep learning in enhancing agricultural productivity and promoting economic growth. The study found that supervised learning networks, such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), AlexNet, and ResNet, are primarily used in agriculture to enhance economic growth. However, there is a need to develop new DL techniques that can improve model performance and reduce inference time for practical applications. In this review, critical research gaps, particularly in the development of new techniques, are analyzed. This study emphasizes the importance of continued research in this area to fully leverage DL's potential of DL for smart farming and to achieve sustainable agricultural development. • Deep learning is vital in agriculture. • Reviewed 129 agriculture based-DL papers. • DL propels smart farming. [ABSTRACT FROM AUTHOR]

Facial reenactment is aimed at animating a source face image into a new place using a driving facial picture. In a few shot scenarios, the present strategies are designed with one or more identities or identity-sustained suffering protection challenges. These current solutions are either developed with one or more identities in mind, or face identity protection issues in one or more shot situations. Multiple pictures from the same entity have been used in previous research to model facial reenactment. In contrast, this paper presents a novel model of one-shot many-to-many facial reenactments that uses only one facial image of a face. The proposed model produces a face that represents the objective representation of the same source identity. The proposed technique can simulate motion from a single image by decomposing an object into two layers. Using bi-layer with Convolutional Neural Network (CNN), we named our model Bi-Layer Graph Convolutional Layers (BGCLN) which utilized to create the latent vector's optical flow representation. This yields the precise structure and shape of the optical stream. Comprehensive studies suggest that our technique can produce high-quality results and outperform most recent techniques in both qualitative and quantitative data comparisons. Our proposed system can perform facial reenactment at 15 fps, which is approximately real time. Our code is publicly available at https://github.com/usaeed786/BGCLN. [ABSTRACT FROM AUTHOR]

Fully Convolution Networks have recently become popular for tackling semantic segmentation problems. However, its performance is dependent on the hyper-parameters it selects, and manually fine-tuning these hyper parameters is a time-consuming task. Hence, in this paper, a hyper-parameter optimized Fully Convolution Encoder Decoder Network (FCEDN) is proposed for dermoscopy image segmentation. The hyper-parameters of the network are optimized by a novel Exponential Neighborhood Grey Wolf Optimization (EN-GWO) algorithm. In EN-GWO, a neighborhood based searching strategy is defined by blending the wolves' individual haunting strategies with their global haunting strategies, emphasizing the right balance between exploration and exploitation. A comprehensive study is conducted using the International Skin Imaging Collaboration (ISIC) 2016 and ISIC 2017 datasets to validate the EN-GWO compared with four variants of GWO, GA, and PSO based hyper-parameter optimization techniques. For the ISIC 2016 and ISIC 2017 datasets, the proposed model can segment skin cancer images with a Jaccard coefficient of 96.41%, 86.85%, Dice coefficient of 98.48%, 87.23%, and accuracy of 98.32%, 95.25% respectively. It is evident from the experimental results that the proposed model outperforms other recent deep learning models such as U-Net, Link-Net, SegNet, and FCN. [ABSTRACT FROM AUTHOR]

Suicidal ideation detection from social media is an evolving research with great challenges. Many of the people who have the tendency to suicide share their thoughts and opinions through social media platforms. As part of many researches it is observed that the publicly available posts from social media contain valuable criteria to effectively detect individuals with suicidal thoughts. The most difficult part to prevent suicide is to detect and understand the complex risk factors and warning signs that may lead to suicide. This can be achieved by identifying the sudden changes in a user's behavior automatically. Natural language processing techniques can be used to collect behavioral and textual features from social media interactions and these features can be passed to a specially designed framework to detect anomalies in human interactions that are indicators of suicidal intentions. We can achieve fast detection of suicidal ideation using deep learning and/or machine learning based classification approaches. For such a purpose, we can employ the combination of LSTM and CNN models to detect such emotions from posts of the users. In order to improve the accuracy, some approaches like using more data for training, using attention model to improve the efficiency of existing models etc. could be done. This paper proposes a LSTM-Attention-CNN combined model to analyze social media submissions to detect any underlying suicidal intentions. During evaluations, the proposed model demonstrated an accuracy of 90.3% and an F1-score of 92.6%, which is greater than the baseline models. [ABSTRACT FROM AUTHOR]

Inspired by image-to-curve transformation and multi-exposure fusion, in this paper, we have developed a new method to treat the low light image enhancement tasks as an extended problem with multiple virtual exposures by a non-linear intensity mapping function. Considering that existing image-to-curve methods have difficulty in obtaining the desired detail and brightness recovery in any one iteration without relying on any ground truth, we propose a virtual multi-exposure fusion strategy to merge the outputs from these different iterations. Specifically, a simple CNN is trained to learn a pixel-wise intensity mapping function and accordingly adjust a given image multiple times. Then the results of all iterations are retained together with the original input image for fusion via a WGIF-based Multi-scale pyramid to obtain a final enhanced output. We present experimental results to demonstrate the effectiveness of the new technique and its state-of-the-art performances. [ABSTRACT FROM AUTHOR]

• We propose a rotation invariant Gabor convolutional neural network (RIGCN). • We learn Gabor-guided convolutional features in the Siamese network architecture. • We explore a convolutional fusion operator to obtain rotation invariant features. • We show the effectiveness of RIGCN for rotation invariant image classification. Gabor filters have been recently integrated with deep convolutional neural networks to learn better features with fewer model parameters. However, during feature learning the rotation invariance is not well addressed. In this paper, we propose a rotation invariant Gabor convolutional neural network (RIGCN) for image classification. First, we transform each input image to generate multiple rotated image instances and feed them into a weight-sharing Siamese network architecture to learn Gabor-guided deep convolutional features. Then, we compute the maximum and average feature responses from all the rotated instances of the same input image and send them into a convolutional fusion module to obtain a rotation invariant image representation. Finally, we use the cross-entropy loss for image classification. In our method, the use of Siamese architecture enables us to obtain rotation invariant features from rotated image instances. The use of convolutional fusion operator enables us to obtain richer statistical features with high efficiency. Experimental results on several benchmark datasets demonstrate the effectiveness of RIGCN for rotation invariant image classification. [ABSTRACT FROM AUTHOR]

• Measured and analyzed the 3D fluorescence spectra of four emulsified oils, revealing their unique fluorescence spectral characteristics. • Proposes an emulsified oil species identification method based on the ResNet50 + HPO-SVM model. • Expansion of emulsified oil fluorescence spectra using CVAE to meet the high data volume requirements of ResNet50. • Compared ResNet50 + HPO-SVM, ResNet50 + PLS-DA, and ResNet50 + KNN methods by statistical parameter. Increased pollution from oil emulsions may cause serious ecological damage. The development of a fast and effective emulsified oil species identification method plays a crucial role in dealing with oil spills on the sea surface. Because of this, this paper proposes an emulsified oil species category identification method based on 3D fluorescence spectroscopy and CNN. First, the 3D fluorescence spectra of emulsified oil were measured by an FLS1000 steady-state fluorescence spectrometer, and their fluorescence characteristics were analyzed in depth. Second, the 3D fluorescence spectral data of emulsified oil was extended using a conditional variational autoencoder (CVAE), and the spectral features were extracted using a pre-trained ResNet50. Finally, the spectral features extracted by ResNet50 were used as inputs to the hunter-prey optimized support vector machine (HPO-SVM), and the results show that the recognition accuracy of this method is 97 %, which is better than other classification models. [ABSTRACT FROM AUTHOR]

In this paper, we propose a technique for hierarchical yoga pose classification (YPC) in a multi-stage multi-tasking framework. We propose a three-stage transfer learning based end-to-end training methodology. Novelty lies in (a) proposed supervised contrastive combined loss function for stage-1 training, (b) proposed Encoder–Decoder network architecture with attention mechanism for stage-3 training, (c) proposed spatial context aware multi-tasking combined loss function for stage-3 training. Firstly, for stage-1 training, we propose the usage of linear combination of three loss functions: cross-entropy, self-supervised contrastive loss and supervised contrastive loss in a multi-tasking manner. We introduce radial and cosine margin into the formulation of self-supervised and supervised contrastive loss to pull feature embeddings of same class closer together compared to feature embeddings of different classes. Weights learned over stage-1 training are subsequently fine-tuned over cross-entropy multi-tasking loss in stage-2. These stage-2 weights are transfer learned and are further fine-tuned in stage-3 training. For stage-3 training, we propose the usage of spatial context aware multi-tasking combined loss function. This loss function leverages on the fine-grained spatial features obtained from HiResCAM. These are processed in parallel with features obtained from XGrad-CAM (sensitivity and conservation axioms satisfying features) to further supervise the learning of hierarchical yoga pose classifier. We exemplify our methodology on the publicly available Yoga-82 large-scale dataset. We report peak Top-1 YPC accuracy of 95.89% over 6 pose classes (Yoga-6), 93.85% over 20 pose classes (Yoga-20) and 90.0% over 82 pose classes (Yoga-82). Our proposed method achieves 6.1% improvement over Top-1 classification accuracy in Yoga-6 hierarchy, 9.3% improvement in Yoga-20 hierarchy and 10.9% improvement in Yoga-82 hierarchy in comparison with state-of-the-art (SOTA) methodology. We achieve the current best Top-1 classification accuracies in all the three YPC hierarchies. [Display omitted] • Novel spatial context aware multi-tasking multi-stage end-to-end training method. • Improved spatial context awareness, evident through improved CAM explanations. • Improved inter class margins and intra class compactness. • Achieved SOTA hierarchical yoga pose classification accuracies on Yoga-82 dataset. [ABSTRACT FROM AUTHOR]

• We proposed a novel method for noise reduction of multichannel heart sound signals, which was based on FastICA algorithm and CNN model. • The method performs classification and consistency analysis of the independent components to extract reliable independent components of heart sounds. • The method improves the SNRs of each channel signal up to 20 dB, which is superior to the Wiener filtering and singular spectrum analysis (SSA). Various noises could contaminate heart sounds and yield inaccurate diagnosis. This paper proposed a framework of heart sound noise reduction using multi-channel structure. Multi-channel heart sound (MCHS) signals were preprocessed, and then decomposed into independent components (ICs) using FastICA algorithm. The ICs were evaluated for consistency using correlation analysis, and reliable ICs were identified. The reliable ICs were classified based on a convolutional neural network (CNN) into two groups: one is of ICs for heart sounds and the other is of ICs for noise, respectively. Consequently, the MCHS signals were reconstructed from the ICs group of heart sounds. To validate the proposed framework, the MCHS signals of 15 subjects were acquired and mixed with breath sounds under different signal-to-noise ratios (SNRs). the simulation results showed that the framework had effective noise reduction performances in suppressing the breath sounds. The SNR of the denoised MCHS could be improved to 20 dB. the proposed framework was better than traditional methods, such as Wiener filter and singular spectrum analysis (SSA). The practical applications proved that the framework is possible to reconstruct the heart sound signals in low signal quality situation. [ABSTRACT FROM AUTHOR]

The escalating prevalence of gastrointestinal disorders, spanning a spectrum from polyps to tumors, underscores the imperative for advanced diagnostic and interventional methodologies. This paper addresses this exigency by introducing telemedicine into the healthcare landscape, presenting an innovative approach for remote patient surgery. Telemedicine harnesses technological advancements to extend the reach of surgical expertise, transcending geographical limitations and augmenting patient-centric care strategies. Within the domain of healthcare 4.0, the integration of Explainable Artificial Intelligence (X-AI) is paramount in the telemedicine sector. This arises from the need to elucidate the decision-making processes of intricate models, particularly in the realm of gastrointestinal disease detection. We introduce TeleXGI , a sophisticated model leveraging Convolutional Neural Network (CNN) architectures ResNet50 and MobileNetV2 with X-AI techniques for accurate gastrointestinal disease classification using the Kvasir-V2 dataset. Advanced X-AI techniques such as Grad-CAM, saliency maps, integrated gradients, attribution heatmaps, and Local Interpretable Model-Agnostic Explanations (LIME) are employed to enhance interpretability. These methodologies unveil the inner workings of CNNs, highlighting impacted regions and augmenting predictive transparency. Our performance assessment validates TeleXGI 's excellent diagnostic accuracy at 98.8% with ResNet50 and 98.5% with MobileNetV2, faster inference with lightweight models, and precise explanation heatmaps essential to the real-time and high-throughput data-processing requirements of Telesurgery. The credibility of the explanation heatmaps is verified against the ground truth masks of Polyps images from the Kvasir-SEG dataset. This research establishes a benchmark for pivotal healthcare applications, showcasing the potential of X-AI in enhancing the interpretability of CNN models. [ABSTRACT FROM AUTHOR]

The aim of data science is to catch up with the data-intensive life style as well as the demand for decision support, which becomes common in various domains such as medical, education and other smart solutions. As such, high quality of data analysis is greatly desired for accurate and effective downstreaming exploitations. This is also true for the domain of astronomical survey like GOTO (Gravitational-wave Optical Transient Observer), where large amount of raw data has been collected daily. This is one of recognised projects that search for transient events with the new breed of optical survey telescopes that can detect the sky faster and deeper. This is accomplished by comparing the night-specific data with the reference such that new bright sources are obtained for further study. However, the huge size of data makes it difficult to sift by naked eyes, thus requiring an automated system. Yet, many conventional machine-learning models have been sub-optimal for this task, as true positives can hardly be recognised due to the nature of imbalance data. This motivates the exploration of convolutional neural networks or CNN for this binary classification problem. Based on existing technologies, the paper reports the original application of basic CNN model to a representative data, which has been designed and generated within the GOTO project. In addition to the improvement over those previous works, this empirical study also includes details of parameter analysis, which will be useful for practice and further investigation. [ABSTRACT FROM AUTHOR]

This paper presents a model for sound classification in construction that leverages a unique combination of Mel spectrograms and Mel-Frequency Cepstral Coefficient (MFCC) values. This model combines deep neural networks like Convolution Neural Networks (CNN) and Long short-term memory (LSTM) to create CNN-LSTM and MFCCs-LSTM architectures, enabling the extraction of spectral and temporal features from audio data. The audio data, generated from construction activities in a real-time closed environment is used to evaluate the proposed model and resulted in an overall Precision, Recall, and F1-score of 91%, 89%, and 91%, respectively. This performance surpasses other established models, including Deep Neural Networks (DNN), CNN, and Recurrent Neural Networks (RNN), as well as a combination of these models as CNN-DNN, CNN-RNN, and CNN-LSTM. These results underscore the potential of combining Mel spectrograms and MFCC values to provide a more informative representation of sound data, thereby enhancing sound classification in noisy environments. • Recognition of various construction activities using audio data inputs. • Combining MFCC-LSTM numerical features with CNN-LSTM spatial maps for enhanced audio representation. • Evaluating model performance on 790 audio samples from four different construction activities. • Comparison of model's performance against five commonly used deep learning and sequential methods. [ABSTRACT FROM AUTHOR]

Recent advancements in Deep Learning models hold the potential to revolutionize the automated analysis of EEG data for early and accurate diagnosis of epileptic seizures. This paper introduces an interpretable hybrid model, integrating Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, with a primary emphasis on two-class epileptic seizure classification. Extensive evaluations across diverse datasets establish the model's resilience and effectiveness. Notably, in the CHB-MIT dataset, the model achieves an average validation accuracy of 92.8 %, an average area under the curve of 93 %, an average specificity of 90.3 %, an average sensitivity of 95 %, an F1-score of 94 %, and an MCC of 88.2 %. In the Siena dataset, an average validation accuracy of 92.7 %, an average area under the curve of 93 %, an average specificity of 84 %, an average sensitivity of 91 %, an F1-score of 92.5 %, and an MCC of 85 % are maintained. In the Helsinki dataset, the model attains an average validation accuracy of 86.4 %, accompanied by an average area under the curve of 86 %, an average specificity of 84 %, a sensitivity of 88 %, an F1-score of 87.8 %, and an MCC of 75.3 %. Furthermore, the proposed model provides a post-hoc explainer utilizing the Shapley Additive Explanations (SHAP) method, specifically the SHAP Gradient Explainer that interprets the predictive model by providing two forms of explanation: (i) Event-wise explanations, elucidating why particular EEG data segments are classified as seizures or non-seizure events, and (ii) Patient-wise explanations that precisely pinpoint the brain lobe and hemisphere responsible for the seizure's origin. The explainer's efficacy is meticulously assessed using ground truth data, yielding localization and lateralization accuracy scores of 85.43 % for the CHB-MIT dataset, 86 % for the Siena dataset, and 79.4 % for the Helsinki dataset. This research contributes to the advancement of the responsible and trustworthy use of Artificial Intelligence in seizure vs. non-seizure EEG classification and interpretation, delivering both precise classification and in-depth explanations. [ABSTRACT FROM AUTHOR]

Text-based person search aims to use text descriptions to search for corresponding person images. However, due to the obvious pattern differences in image and text modalities, it is still a challenging problem to align the two modalities. Most existing approaches only consider semantic alignment within a global context or partial parts, lacking consideration of how to match image and text in terms of differences in model information. Therefore, in this paper, we propose an efficient Modality-Aligned Person Search network (MAPS) to address this problem. First, we suppress image-specific information by image feature style normalization to achieve modality knowledge alignment and reduce information differences between text and image. Second, we design a multi-granularity modal feature fusion and optimization method to enrich the modal features. To address the problem of useless and redundant information in the multi-granularity fused features, we propose a Multi-granularity Feature Self-optimization Module (MFSM) to adaptively adjust the corresponding contributions of different granularities in the fused features of the two modalities. Finally, to address the problem of information inconsistency in the training and inference stages, we propose a Cross-instance Feature Alignment (CFA) to help the network enhance category-level generalization ability and improve retrieval performance. Extensive experiments demonstrate that our MAPS achieves state-of-the-art performance on all text-based person search datasets, and significantly outperforms other existing methods. • A novel text-based person search network is proposed by reducing modal differences while learning sufficient modal features. • A multi-granularity feature self-optimization module is designed to optimize the multiscale image modal feature and multi-level semantic text modal feature, so as to learn more discriminative features with suppressing useless and redundant information. • A cross-instance feature alignment is proposed to construct image–text feature pairs with category-level information participating in training. • Extensive experiments in both CUHK-PEDES and ICFG-PEDES datasets show our MAPS obtains the state-of-the-art performance, which significantly outperforms other existing methods. [ABSTRACT FROM AUTHOR]

• A lightweight model for pig face recognition was developed using a fourth-order Runge-Kutta numerical method and a hybrid attention module. Subsequent experiments were conducted to evaluate its performance. • The model effectively enhances training and inference efficiency, reduces the number of parameters and memory usage, and improves accuracy in the extraction of pig face information. • Experimental tests show that the model achieves 99.26 % recognition accuracy and has a size of only 1.52 MB, which makes it suitable for deployment on embedded devices. • The heatmap visualization of the model's attention to the region of interest shows that the model pays more attention to the pig face region compared to other models. • The model can provide technical support for the scientific management of pig farms, thus improving the feasibility of deploying lightweight intelligent devices in smart pig farms. Pig face recognition plays a significant role in intelligent pig farming. Accurate and lightweight methods for recognizing pig faces are crucial for precision pig farming. Generally, under the condition of the same computing resources, the lower the image resolution, the faster the model inference speed. Based on the above idea, this paper proposes a lightweight pig face model called RKNet-HAM, based on the fourth-order Runge-Kutta and hybrid attention mechanism. This model exhibits the most prominent focus on the semantic information of low-resolution(64 × 64) pig face images and achieves high accuracy in recognizing similar pig faces with a low error classification rate. A dataset of 21,742 images of 32 pigs was constructed for individual pig recognition. The proposed RKNet-HAM model and several other deep learning models, e.g., RKNet-CBAM, RKNet, ShuffleNetV2.0, DesNet121, and ResNet50. RKNet-HAM has a size of 1.52 Megabytes. Experimental results show that among these models, the proposed model ofachieved the highest accuracy, precision, recall, and specificity, with an accuracy rate of 99.26 %. RKNet-HAM also exhibits good generalization ability. It provides experimental support for mobile and embedded applications. [ABSTRACT FROM AUTHOR]

• Interpretable network design: NEFGN incorporates physical models to make the denoising network more interpretable. By grounding the network design in physical principles, it becomes easier to understand the underlying processes of noise reduction. • The NEFGN method proposed in this paper combines the advantages of traditional denoising and deep learning denoising to provide better interpretability and stronger noise reduction ability. • Construction of noise optimization model stage: By learning the parameters of the adaptive weighted total variation regularization model, NEFGN can approximate the noise level in NDCT images, so as to remove the noise in LDCT images more accurately. In addition, the study introduced noise level estimation maps as a means to accurately control the balance between noise reduction and detail retention for optimal visualization and image quality. • Construction of Gaussian curvature edge detection model stage: NEFGN uses Gaussian curvature guidance to construct an edge detection network to directly predict clear edges from LDCT images. This helps to preserve important edge details and avoid image blurring or information loss. And an edge detector for making edge labels is proposed, which can better deal with the noise reduction task of low-dose CT images. • Comprehensive experimental evaluation: The proposed method was evaluated using the "NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge" dataset, piglet dataset, and clinical dataset. The experimental results demonstrate that NEFGN effectively enhances the diagnostic performance of CT images by restoring the LDCT image structure with constrained image details. • Superior performance: NEFGN outperforms other denoising algorithms in terms of both quantitative measures and visual quality. This makes it a promising candidate for clinical applications, where accurate interpretation of CT images is crucial for diagnosis and treatment planning. The denoising task of low-dose CT images is a highly complex and uncertain inverse problem. Previous studies have primarily relied on convolutional neural network to reduce noise by learning the mapping from LDCT images to normal dose CT images. However, simply increasing the network depth alone is not an optimal choice due to the limited performance improvement and significant computational cost. In contrast, integrating prior knowledge of images with a model to assist in image reconstruction is a more efficient approach. This study proposes a new framework for denoising LDCT images, named Noise-Optimized Edge Feature Guided Network (NEFGN). The task of NEFGN is to integrate the noise optimization model of adaptive weighted total variation expansion, the edge detection model guided by Gaussian curvature, and image reconstruction into an end-to-end CNN framework. In order to achieve this goal, the noise optimization model is first constructed by learning the parameters in the adaptive weighted total variation regularization model to approximate the noise level in the NDCT image. The edge detection network is constructed using Gaussian curvature, predicting clear edges directly from the noise image. Finally, under the guidance of the noise optimization model and the edge detail model, NEFGN is more capable of suppressing artifact noise, demonstrating good accuracy and robustness, and can restore finer details. Numerous experimental studies demonstrate that the NEFGN denoising framework effectively restores the structure of LDCT images with limited image details and outperforms other methods in terms of performance. [ABSTRACT FROM AUTHOR]

This paper demonstrates L-band quantum-dash laser (QDL) based orbital angular momentum (OAM) structured light in a free space optics communication (FSO) system. A 4-ary OAM-shift-keying pattern coding communication system, based on Laguerre Gaussian (LG) and superposition LG (MuxLG) mode families, has been investigated under a foggy FSO channel. In addition, joint mode identification and channel condition estimation have been developed at the receiver side using advanced deep learning (DL) methods. We utilize and compare the performance of the convolutional neural networks (CNN) and UNET algorithms. An experimental setup has been conducted using an in-house controlled foggy chamber which allows an FSO transmission of 3-m length. Furthermore, we propose a data balancing approach to the experimental dataset by data augmentation. Visibility prediction results have shown a measured root mean square error of 17(18) m and 10(10) m for 4-ary LG(MuXLG) using CNN and UNET models, respectively. Moreover, the DL models provide an average mode classification accuracy of 94% under various channel visibility conditions. • OAM mode patterns was generated using QDL source and identified under foggy channel conditions. • Eight modes have been formed from the LG mode family. • A controlled chamber environment to emulate the outdoor fog condition was considered. • CNN and UNET DL methods were used as classifiers and regressors for mode identification and channel condition prediction. • Visibility parameter can be estimated with 17 and 10 m root mean square error (RMSE) using CNN and UNET techniques. [ABSTRACT FROM AUTHOR]

A tunnel boring machine (TBM) generates rock chips during excavation, which are crucial for assessing surrounding rock integrity, enhancing excavation efficiency, and evaluating cutter wear. However, traditional methods struggle to identify small rock chips, chips submerged in soil or water, and chips in stacked states. This paper proposes a convolutional neural network (CNN)-based method for directly recognizing the particle size distribution from rock chip images. A dataset of 2520 rock chip images representing 84 particle-size distributions was collected in a laboratory environment. By comparing various CNN architectures and hyperparameters, an optimal model was obtained with a mean absolute error (MAE) of 1.66 × 10−2 and R 2 of 0.923 on the test set. The results demonstrate that the proposed method enables the real-time recognition of particle size distribution using rock chip images, which has the potential to significantly improve intelligent auxiliary excavation technology in TBMs. • A dataset of 2520 rock chip images representing 84 particle size distributions was collected. • A CNN-based method for directly recognizing particle size distribution from rock chip images is proposed. • An optimal model was obtained with a MAE of 1.66 × 10−2 and R 2 of 0.923 on the test set. [ABSTRACT FROM AUTHOR]

In today's digital age, the proliferation of network-connected devices has triggered a surge in cyberattacks. Distributed Denial-of-Service (DDoS) attacks pose a particularly formidable challenge to network security by disrupting access to vital services. While numerous researchers have proposed DDoS detection methods utilizing machine learning and deep learning techniques, developing a robust and reliable DDoS intrusion detection system remains challenging. This challenge is exacerbated by issues such as highly imbalanced data, multi-classification, and computational complexity. This paper proposes an innovative feature selection approach to create a robust intrusion detection system capable of detecting and classifying recent common DDoS attack types. We evaluate the performance of our model on the CICDDoS2019 benchmark dataset. Our experimental results demonstrate that our proposed model outperforms existing methods, achieving a detection accuracy of 96.82%, a recall of 96.82%, a precision of 96.76%, and an F1 score of 96.50%. Additionally, our model exhibits faster prediction times, with the ability to predict an attack in just 0.189 ms. Notably, our approach, combined with preprocessing and feature selection techniques, outperforms previous works and baseline models in DDoS attack classification. [Display omitted] • An innovative hybrid feature extraction method for training of DDoS attack detection model. • We propose a robust CNN model integrated with an Inception mechanism designed to classify various types of DDoS attacks. • Evaluated the network performance of the proposed system in terms of various parameters and compared the results with existing studies and baseline models. • The proposed model outperforms more than 96% in all performance metrics (Precision, Recall, Accuracy, and F1 Score). • The proposed model achieves the fastest prediction time of 0.189 ms among existing models in the literature. [ABSTRACT FROM AUTHOR]

MeV Gamma-ray Telescope (MGT) is a conceptual mission aimed at improving the detection sensitivity of gamma-ray astronomy in the MeV energy range. It consists of three sub-detectors: Gamma-ray Conversion silicon tracker, CALOrimeter and Anti-Coincident Detector. In this paper, a track reconstruction algorithm based on Convolutional Neural Networks (CNN) is developed for MGT. In order to train and test the model, Geant4 simulation is used and generates a large number of gamma-ray events at nine energy points in the energy band from 0.5 GeV to 10 GeV. Finally, the reconstruction results of angular resolution, position resolution and acceptance are shown. The testing results indicate that the angular resolution of MGT significantly improves in the 0. 5 ∼ 10 GeV range compared with Fermi-LAT. [ABSTRACT FROM AUTHOR]

The Brain Tumor is one of the most serious scenarios associated with the brain where a cluster of abnormal cells grows in an uncontrolled fashion. The field of image processing has experienced remarkable growth in the area of biomedical applications with the invention of different techniques in deep learning. Brain tumor classification and detection is a subject of prime importance where Convolutional Neural Networks (CNN) find application. But the main drawback of the existing technology is that it is complex with a huge number of parameters contributing to high execution time and high system specifications for implementation. In this paper, a novel architecture for Brain tumor classification and tumor type object detection using the RCNN technique is proposed which has been analyzed using two publicly available datasets from Figshare (Cheng et al., 2017) and Kaggle (2020). Here we aim to decrease the execution time of a conventional RCNN architecture with the use of a low complex framework and propose a system for brain tumor analysis. We first use a Two Channel CNN, a low complex architecture to classify between Glioma and healthy tumor MRI samples which was successfully done with an accuracy of 98.21 percentage. Later this same architecture is used as the feature extractor of an RCNN to detect the tumor regions of the Glioma MRI sample that has been classified from the previous stage and the tumor region is bounded using bounding boxes. Also, this method has been extended to other two types of tumors Meningioma and Pituitary tumor. The methodology was able to achieve very low execution time as compared with the other existing architectures with an average confidence level of 98.83 percentage. [ABSTRACT FROM AUTHOR]

CNNs with strong learning abilities are widely chosen to resolve super-resolution problem. However, CNNs depend on deeper network architectures to improve performance of image super-resolution, which may increase computational cost in general. In this paper, we present an enhanced super-resolution group CNN (ESRGCNN) with a shallow architecture by fully fusing deep and wide channel features to extract more accurate low-frequency information in terms of correlations of different channels in single image super-resolution (SISR). Also, a signal enhancement operation in the ESRGCNN is useful to inherit more long-distance contextual information for resolving long-term dependency. An adaptive up-sampling operation is gathered into a CNN to obtain an image super-resolution model with low-resolution images of different sizes. Extensive experiments report that our ESRGCNN surpasses the state-of-the-arts in terms of SISR performance, complexity, execution speed, image quality evaluation and visual effect in SISR. Code is found at https://github.com/hellloxiaotian/ESRGCNN. [ABSTRACT FROM AUTHOR]

In this paper, a clustering method for KMedoids based on dynamic time warping (DTW-KMedoids) is designed to analyze multi-channel signals, and a lightweight network, clustered blueprint separable convolutional neural network (CBS-CNN), is established to perform fault diagnosis of high-speed train (HST) bogie. The motivation for proposing the novel method is to address the problems of large network size and high training cost in deep learning. First, DTW-KMedoids is adopted to cluster the channels of multi-channel signals. Second, based on the principles of blueprint separable convolution (BSConv) and mixed depthwise convolution (MixConv), a lightweight convolution model construction strategy called clustered blueprint separable convolution (CBS-Conv) is proposed, which uses the same blueprint to convolute the data of the channels in a cluster. Third, CBS-CNN is established, with multiple branches to process data from different clusters, and the computational result of each branch is connected by the proposed Connect layer. Finally, by virtue of the learned features from training, the model completes the end-to-end HST bogie fault diagnosis task, where the usefulness of CBS-CNN in detecting bogie failures including component performance degradation, component failures, and composite failures is validated. Further experiments show that CBS-CNN has a remarkable ability to adapt itself to different task environments and objectives. [ABSTRACT FROM AUTHOR]

As a result of late diagnosis, cancer is the second leading cause of death in most countries in the world. Usually, many cases of cancer are diagnosed at an advanced stage, which reduces the chances of recovery from the disease due to the inability to provide appropriate treatment. The earlier cancer is detected, the more effective the treatment can be, especially for incurable cancers, which can result in a shorter life expectancy due to the rapid spread of the disease. The early detection of cancer also greatly reduces the financial consequences of it, as the cost of treating it in its early stages is much lower than in its other stages. Therefore, several previous studies focus on developing computer-aided cancer diagnosis systems (CACDs) that can detect cancer in its earliest stages automatically. In this paper, a novel approach is proposed for cancer detection. The proposed approach is an end-to-end deep learning approach, where the input images are fed directly to the deep model for final decision. In this research, the accuracy of a new deep convolutional neural network (CNN) for cancer detection is explored. The microscopic medical images obtained from the cancer database were used to evaluate our study, which were labelled as normal and abnormal images. The presented model achieved an accuracy of 99.99%, which is the highest accuracy compared with other deep learning models. Finally, the proposed approach would be very useful and effective, especially in low-income countries where referral systems for patients with suspected cancer are often unavailable, resulting in delayed and fragmented care. [ABSTRACT FROM AUTHOR]

The best scheduling of water distribution systems may be supported by accurate and trustworthy water demand forecasts, which is a positive assurance for the development of smart cities and smart water services. This paper studies a new multivariate time-series prediction model that is based on the Convolution Neural Network (CNN) and Gate Recurrent Unit(GRU), taking into account the limitations of the prediction of multivariate time series using a single model. According to the regularity of water use during the time period, the characteristics of water users are clustered and classified to form two types of users: tidal type and irregular type. The factors that affect the daily water consumption such as user type, week, water consumption of the day of the previous three weeks, major events, etc. are used as input vectors, and the 2018-2021 resident daily water consumption time series data of a domestic water company is used as the training sample, respectively establish CNN-GRU models.CNN-GRU approach is checked using the root mean square error(RMSE),mean absolute error(MAE),and mean percentage absolute error(MAPE).The results are compared with Long-Short Term Memory (LSTM),CNN and GRU.Results show that CNN-GRU improves water demand prediction. The CNN-GRU model's RMSE dropped by around 0.648, 0.82, 0.82 when compared to the LSTM, CNN, and GRU models.the MAE decreased by about 0.418,0.47,0.462;and the MAPE decreased by about 0.722,0.649,0.712. [ABSTRACT FROM AUTHOR]

UAV (Unmanned Aerial Vehicle) monitoring mounted with high resolution camera is a rising way to monitor the ocean environment, and it can make up the shortages of low spatial and temporal resolutions of SAR images. How to get the accurate regions of Ulva prolifera in the very high-resolution images remains a lot of challenges. Due to the limitation of GPU memory, the popular pixel-level image segmentation methods cannot deal with the raw resolution images(Up to 6000*4000). In this paper, we propose a novel framework to get the Ulva prolifera regions, which incorporates both superpixel segmentation and CNN classification and can deal with raw resolution images. We first process the raw images with superpixel algorithm to generate local multi-scale patches. And then a binary classification CNN model can be trained with the labeled patches. With the result of superpixel segmentation and the classification of CNN model, a more detailed segmentation of Ulva prolifera can be obtained. Two datasets UlvaDB-1 and UlvaDB-2 are also proposed in this paper. The experiment results show that the proposed method can achieve state-of-the-art performance compared with the recent pixel-level segmentation and instance-aware semantic segmentation methods. [ABSTRACT FROM AUTHOR]

Product service system (PSS), an undeveloped business model, provides consumers with full-life cycle services from the perspective of products, functions and services. This paper proposes a posture recognition method based on smart product service system from the aspects of technology and user' need. A Posture-CNN model based on the convolutional neural network (CNN) is established in this paper. According to the theory of PSS, this paper analyses the difficulties that the posture recognition technology is facing in the application of products, and resolves the existing problems such as low accuracy of recognition and unsupported complex environment. A self-made posture data set is tested with the results that the method can greatly reduce network parameters and improve network speed compared with other existing classification methods. To some extent, as a core technology of the PSS control terminal, this technology can perfect users' experiences, better serve consumers and promote the development of the PSS. [ABSTRACT FROM AUTHOR]

Highlights • Reports on the use of a self-learning softmax with a 9-layer CNN model to realize nighttime pedestrian recognition. • 60,000 Near-Infrared (NIR) images were collected on the urban and rural road by a car-mounted NIR camera with a visible-spectrum filter and NIR fill-light. • Testing accuracy reached 94.49% on a set of 15,000 complex testing samples. On a desktop with a common GPU, the single sample recognition time was only 0.07 ms. Abstract Pedestrian recognition is the core technology of pedestrian detection in pedestrian protection systems. This paper compares and analyzes, visible and infrared images obtained via visible-spectrum, near-infrared, short-wave infrared, and long-wave infrared cameras. The results show that near-infrared camera was the best for nighttime pedestrian detection when device cost and pedestrian imaging quality were considered. This paper reports on the first time use of a self-learning softmax with a 9-layer Convolutional Neural Network (CNN) model to identify near-infrared nighttime pedestrians. 267,000 samples obtained from the near-infrared images were employed to optimize the CNN recognition model. Collected near-infrared nighttime samples had 3 categories (background, pedestrian, and cyclist or motorcyclist) and will be made publicly available for researchers use. Testing results indicated that the optimized CNN model using self-learning softmax had a competitive accuracy and potential in real-time pedestrian recognition. [ABSTRACT FROM AUTHOR]

Abstract For autonomous driving system, vehicle detection is an import part as well as a challenging problem due to the large intra-class differences caused by occlusion, truncation and different viewpoints. The detection system should be fast and accurate enough to support real-world applications. Most of the existing deep convolution neural network (CNN) based object detection methods can be roughly categorized into two streams: single-stage and two-stage modes. These single-stage methods are usually extremely fast and easy to train with losing some precision. As for two-stage methods, they often get high performance in object detection competitions, however, they are not competitive for real-world applications because of the speed limits. The detection system with high degree of precision and fast computation speed is desirable. In this paper, a new two-stage regression based cascade object detection system is proposed. This system can be fast detection of the vehicles which concentrated the advantages of the two aforementioned methods, denoted by HybridNet. In our design, the first and the second stage are both regression modes. We add a transitional stage to map proposals (generated in the first stage) on high resolution feature maps to get exact features for decision refinement in the second stage. The challenging KITTI and PASCAL VOC2007 data sets are used to evaluate our proposed method. The experimental results show that our approach is more fast and more accurate in vehicle detection than other state-of-the-art methods. Highlights • In this paper, a new two-stage regression based cascade object detection system is proposed. • This system can be fast detection of the vehicles which concentrated the advantages of the single-stage and two-stage methods, denoted by HybridNet. • In our design, the first and the second stage are both regression modes. • We add a transitional stage to map proposals(generated in the first stage) on high resolution feature maps to get exact features for decision refinement in the second stage. • The challenging KITTI and PASCAL VOC2007 data sets are used to evaluate our proposed method. • The experimental results show that our approach is more fast and more accurate in vehicle detection than other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Segmentation of medical images is a critical step to distinguish different regions, especially tumor. The presence of severe intensity inhomogeneity in Magnetic Resonance Imaging (MRI) and Mammogram images makes it difficult to assess and detect the boundaries of tumor. Tumor detection in medical images measures and identifies the structure of the tumor. This challenging task of extracting tumor i.e., region of interest is achieved by proposing a novel fuzzy selective model. The proposed model is mainly based on selective segmentation to detect the tumor in medical images. This model is based on the local Gaussian distribution as a fitting term with fuzzy logic to handle uncertainty and irregular borders in medical images. Due to the usage of fuzzy membership functions, it avoids local minima and converges to the tumor boundary rapidly and efficiently. In addition, to obtain more information of a local intensity, the idea of multi-scale modeling is incorporated to achieve significant results. The proposed segmentation model is applied to diverse MRI and Mammogram datasets containing both the intensity inhomogeneity and noisy images, this data is collected from local hospitals and the ground truths are created by the experts. The extensive simulation results show that the proposed segmentation method significantly outperforms existing state-of-the-art techniques using both the qualitative and quantitative assessment. Loss functions play very important role in CNN architectures, and is well studied research area. In the second part of the paper, we will be introducing a new loss function, which is introduced in the first part of the paper. Deep active contour architectures based Chan-Vese loss function works well in images with homogeneity and may not work well in images with intensity inhomogeneity. We proposed an architecture, whose loss function will be based on this new proposed model. The proposed model is used as a loss function in a newly introduced CNN architecture, which has produced improved results. The results of the new architecture are compared with other existing methods used for segmentation of medical images. • A new model based on multi-scale image data is proposed for detection of tumor. • The model is further improved by using local Gaussian distribution combined with fuzzy set framework. • A new CNN based architecture based on Res-Net50 and DeepLab v3+ integrated with the new loss function is proposed. • Testing is done on local collected datasets. [ABSTRACT FROM AUTHOR]

Non-Small Cell Lung Cancer is the most common type of lung cancer, accounting for more than 80% of all cases. The analysis of histopathological images is the appropriate way to detect lung cancer in its initial stages. However, due to the heterogeneous patterns of molecular structures and the high magnification factor, the classification of whole slide images is difficult. Furthermore, these large-pixel images cannot be considered directly due to the high computational requirements. Several deep-learning models have been developed for the classification of histopathological images, but only in the default RGB format. Existing models are complex and rely on pathologist annotations which increases the computational cost and time. The present paper proposes a deep-learning framework, a Color-based Dilated Convolutional Neural Network (CD-CNN) for the multi-class classification of patch-based whole slide images of lung cancer. The proposed CD-CNN framework includes data preparation, color space transformation, and the classification stages. It thoroughly examines the impact of five color spaces on classification performance. The proposed model was also compared to three pre-trained models and the Convolutional Neural Network model on three different datasets namely: The Cancer Genome Atlas dataset, the Clinical Proteomic Tumor Analysis Consortium cohorts, and the LC25000 dataset. The analysis of all the models reveals that the HSV color space outperforms the other transformations for all of the datasets tested. The results depict that the proposed CD-CNN achieved the best classification results in HSV color space for all three datasets i.e. accuracy of 0.97–0.99, precision, recall, and F1 score 0.97–0.98 and AUC value of 0.970–0.984. In addition, the highest value of the kappa score 0.986 has been achieved by the proposed model. These findings demonstrate the efficacy of the proposed method and show an improvement in diagnostic systems. • The analysis of Whole Slide Images (WSIs) for the classification of lung cancer is difficult due to its heterogeneous structure. • The present paper proposes a color-based deep-learning framework for the classification of lung cancer-type WSIs. • The proposed framework thoroughly investigates the impact of five color spaces: RGB, YCbCr, LAB, HSV, and XYZ on the classification performance. • Three different lung cancer datasets have been used to validate and test the performance of the proposed model. Thorough investigation and comparative analysis indicate the efficacy of the proposed framework over other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Dysarthria is a motor speech disability caused by weak muscles and organs involved in the articulation process, thereby affecting the speech intelligibility of individuals. Because this condition is linked to physical exhaustion disabilities, individuals not only have communication difficulties, but also have difficulty interacting with digital devices. Automatic speech recognition (ASR) makes an important difference for individuals with dysarthria since modern digital devices offer a better interaction medium that enables them to interact with their community and computers. Still, the performance of ASR technologies is poor in recognizing dysarthric speech, particularly for acute dysarthria. Multiple challenges, including dysarthric phoneme inaccuracy and labeling imperfection, are facing dysarthric ASR technologies. This paper proposes a spatio-temporal dysarthric ASR (DASR) system using Spatial Convolutional Neural Network (SCNN) and Multi-Head Attention Transformer (MHAT) to visually extract the speech features, and DASR learns the shapes of phonemes pronounced by dysarthric individuals. This visual DASR feature modeling eliminates phoneme-related challenges. The UA-Speech database is utilized in this paper, including different speakers with different speech intelligibility levels. However, because the proportion of usable speech data to the number of distinctive classes in the UA-speech database was small, the proposed DASR system leverages transfer learning to generate synthetic leverage and visuals. In benchmarking with other DASRs examined in this study, the proposed DASR system outperformed and improved the recognition accuracy for 20.72% of the UA-Speech database. The largest improvements were achieved for very-low (25.75%) and low intelligibility (33.67%). [ABSTRACT FROM AUTHOR]

Aspect-based sentiment analysis (ABSA) may provide more detailed information than general sentiment analysis. It aims to extract aspects from reviews and predict their polarities. In this paper, we focus on aspect extraction sub-task. We propose three weakly-supervised systems based on contextual language models and topic modeling. We evaluate and compare our systems on SemEval-2016 restaurant french benchmark. The experimental results reveal that our systems is quite competitive in aspect extraction from user reviews. We obtain 60.65% as F1 score with our best system. The latter outperforms the existing supervised ones. We deduct that weakly-supervised systems are efficient in terms of performance, time and human effort. [ABSTRACT FROM AUTHOR]

Musical instrument recognition is an important task within the broad field of Music Information Retrieval. It helps to build recommendation systems, compute similarity between musical compositions and enable automatic search in music collections with regard to the instrument. The task has two variants differed by the difficulty level. The simpler one is classification based on the sound of a single instrument, while the more difficult challenge is to recognize the predominant instrument in polyphonic recordings. In this paper, we used a convolutional neural network to solve both of these problems. As the analysis of monotimbral recordings is relatively easy, we used the knowledge acquired during solving it to train a neural network to tackle the more complex predominant instrument recognition problem. Within this staged training scenario, we also examined the impact of introducing some intermediate stages during the training sessions. The results showed that such a training approach has a potential to improve the accuracy of classification. [ABSTRACT FROM AUTHOR]

The paper examines the accuracy of classification performed by selected deep learning methods on a large number of ECG data streams. Specifically, it investigates the performance of two types of neural networks: the convolutional neural network (CNN) and residual neural network (ResNet). The results were compared with those obtained by the classic k-nearest neighbors (k-NN) approach. A number of significant signal and method-specific parameters were taken into account, like the duration of the signal examined, the distance between consecutive R waves, the analysis window, the number of learning epochs, etc. The investigation was performed on two data sets: PTB and MIT (Arrhythmia). [ABSTRACT FROM AUTHOR]

Recommendation systems (RS) play an important role in e-commerce applications as they help the consumers in choosing the required items within reduced time. The traditional methods of collaborative filtering, fail to capture the visual data associated with the items. Visually-aware recommendation systems are upcoming in e-commerce applications that use the visual features of the products rather than the user profiles. Deep learning techniques are used for the classification and prediction in visual recommendation systems. However, the criticality of the visual recommendation system lies in identifying the similar images for a given target image. In this paper, a visual recommendation system is proposed based on Deep Visual Ensemble similarity metric (DVESM) using Convolutional Autoencoder (CAE) neural network for classification. The basic idea is to get a set of trained feature vectors for the image catalogue using CAE and find the similarity between the trained feature vectors and the target feature vector using DVESM method. The proposed methodology of using DVESM has been demonstrated with the state-of-the-art methods on Amazon 2014, 2015 and Street2Shop datasets. The results show that the DVESM method is most suitable for visually aware recommendation systems as it learns an ensemble of metrics to provide recommendations. [ABSTRACT FROM AUTHOR]