Search

Single-frame infrared small target detection is affected by the low image resolution and small target size, and is prone to the problems of small target feature loss and positional offset during continuous downsampling; at the same time, the sparse features of the small targets do not correlate well with the global-local linkage of the background features. To solve the above problems, this paper proposes an efficient infrared small target detection method. First, this paper incorporates BlurPool in the feature extraction part, which reduces the loss and positional offset of small target features in the process of convolution and pooling. Second, this paper designs an interactive attention deep feature fusion module, which acquires the correlation information between the target and the background from a global perspective, and designs a compression mechanism based on deep a priori knowledge, which reduces the computational difficulty of the self-attention mechanism. Then, this paper designs the context local feature enhancement and fusion module, which uses deep semantic features to dynamically guide shallow local features to realize enhancement and fusion. Finally, this paper proposes an edge feature extraction module for shallow features, which utilizes the complete texture and location information in the shallow features to assist the network to initially locate the target position and edge shape. Numerous experiments show that the method in this paper significantly improves nIoU, F1-Measure and AUC on IRSTD-1k Datasets and NUAA-SIRST Datasets. [ABSTRACT FROM AUTHOR]

In industry, the task of defect classification and defect localization is an important part of defect detection system. However, existing studies only focus on one task and it is difficult to ensure the accuracy of both tasks. This paper proposes a defect detection system based on improved Yolo_v4, which greatly improves the detection ability of minor defects. For K_Means algorithm clustering prianchors question with strong subjectivity, the paper proposes the Density Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to determine the number of Anchors. To solve the problem of low detection rate of small targets caused by insufficient reuse rate of low-level features in CSPDarknet53 feature extraction network, this paper proposes an ECA-DenseNet-BC-121 feature extraction network to improve it. And the Dual Channel Feature Enhancement (DCFE) module is proposed to improve the local information loss and gradient propagation obstruction caused by quad chain convolution in PANet networks to improve the robustness of the model. The experimental results on the fabric surface defect detection datasets show that the mAP of the improved Yolo_v4 is 98.97%, which is 7.67% higher than SSD, 3.75% higher than Faster_RCNN, 10.82% higher than Yolo_v4 tiny, and 5.35% higher than Yolo_v4, and the detection speed reaches 39.4 fps. It can meet the real-time monitoring needs of industrial sites. [ABSTRACT FROM AUTHOR]

Introduction: Chemical special steels are widely used in chemical equipment manufacturing and other fields, and small defects on its surface (such as cracks and punches) are easy to cause serious accidents in harsh environments. Methods: In order to solve this problem, this paper proposes an improved defect detection algorithm for chemical special steel based on YOLOv8. Firstly, in order to effectively capture local and global information, a ParC2Net (Parallel-C2f) structure is proposed for feature extraction, which can accurately capture the subtle features of steel defects. Secondly, the loss function is adjusted to MPD-IOU, and its dynamic non-monotonic focusing characteristics are used to effectively solve the overfitting problem of the bounding box of low-quality targets. In addition, RepGFPN is used to fuse multi-scale features, deepen the interaction between semantics and spatial information, and significantly improve the efficiency of cross-layer information transmission. Finally, the RexSE-Head (ResNeXt-Squeeze-Excitation) design is adopted to enhance the positioning accuracy of small defect targets. Results and discussion: The experimental results show that the mAP@0.5 of the improved model reaches 93.5%, and the number of parameters is only 3.29M, which realizes the high precision and high response performance of the detection of small defects in chemical special steels, and highlights the practical application value of the model. The code is available at https://github.com/improvment/prs-yolo. [ABSTRACT FROM AUTHOR]

Emotion recognition plays an increasingly important role in today's society and has a high social value. However, current emotion recognition technology faces the problems of insufficient feature extraction and imbalanced samples when processing speech and text information, which limits the performance of existing models. To overcome these challenges, this paper proposes a multi-modal emotion recognition method based on speech and text. The model is divided into two channels. In the first channel, the extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS) feature set is extracted from OpenSmile, and the original eGeMAPS feature set is merged with the wavelet transformed eGeMAPS feature set. Then, speech features are extracted through a sparse autoencoder. The second channel extracts text features through the BERT-RoBERTa model. Then, deeper text features are extracted through a gated recurrent unit (GRU), and the deeper text features are fused with the text features. Emotions are identified by the attention layer, the dual-layer Bidirectional Long Short-Term Memory (BiLSTM) model, and the loss function, combined with cross-entropy loss and focus loss. Experiments show that, compared with the existing model, the WA and UA of this model are 73.95% and 74.27%, respectively, on the imbalanced IEMOCAP dataset, which is superior to other models. This research result effectively solves the problem of feature insufficiency and sample imbalance in traditional sentiment recognition methods, and provides a new way of thinking for sentiment analysis application. [ABSTRACT FROM AUTHOR]

In order to solve the related problems of detection and tracking of shallow marine organisms, this paper designs a YOLO v5 multi‐target detection and tracking algorithm with attention mechanism. When working underwater, the authors usually encounter many difficulties. Different luminosity and complex coral background usually affect the detection of marine organisms. At the same time, the unrestricted movement of marine organisms, the ability to hide behind rocks and algae, and their mutual occlusion while swimming pose additional challenges to this task. Considering the characteristics of shallow marine organisms activity environment, the attention mechanism is added to the feature extraction network of YOLO v5 to reduce redundant information and improve the detection accuracy of shallow marine organism targets in complex environment. The improved algorithm improves the average detection accuracy of marine organisms target detection by 3.2%. Aiming at the problem of shallow marine organisms target tracking, a shallow marine organisms multi‐target tracking algorithm based on improved Deep Simple Online And Realtime Tracking (SORT) is designed. The improved YOLO v5 algorithm is used to replace Faster R‐CNN (Faster Region‐Convolutional Neural Networks) as the detector of DeepSORT tracking algorithm, and the cascade matching strategy is adopted to solve the problem that the target cannot be tracked continuously when it is occluded for a long time. The experimental results show that the shallow marine organisms multi‐target tracking algorithm based on improved DeepSORT reduces the number of id transformation of marine biological target tracking in shallow sea environment, and improves the accuracy of shallow marine organisms multi‐target tracking. [ABSTRACT FROM AUTHOR]

It is difficult to visually detect internal defects in a large transformer with a metal closure. For convenient internal inspection, a micro-robot was adopted, and an inspection method based on an image-enhancement algorithm and an improved deep-learning network was proposed in this paper. Considering the dim environment inside the transformer and the problems of irregular imaging distance and fluctuating supplementary light conditions during image acquisition with the internal-inspection robot, an improved MSRCR algorithm for image enhancement was proposed. It could analyze the local contrast of the image and enhance the details on multiple scales. At the same time, a white-balance algorithm was introduced to enhance the contrast and brightness and solve the problems of overexposure and color distortion. To improve the target recognition performance of complex carbon-trace defects, the SimAM mechanism was incorporated into the Backbone network of the YOLOv8 model to enhance the extraction of carbon-trace features. Meanwhile, the DyHead dynamic detection Head framework was constructed at the output of the YOLOv8 model to improve the perception of local carbon traces with different sizes. To improve the defect target recognition speed of the transformer-inspection robot, a pruning operation was carried out on the YOLOv8 model to remove redundant parameters, realize model lightness, and improve detection efficiency. To verify the effectiveness of the improved algorithm, the detection model was trained and validated with the carbon-trace dataset. The results showed that the MSH-YOLOv8 algorithm achieved an accuracy of 91.80%, which was 3.4 percentage points higher compared to the original YOLOv8 algorithm, and had a significant advantage over other mainstream target-detection algorithms. Meanwhile, the FPS of the proposed algorithm was up to 99.2, indicating that the model computation and model complexity were successfully reduced, which meets the requirements for engineering applications of the transformer internal-inspection robot. [ABSTRACT FROM AUTHOR]

With the widespread use of UAVs, UAV aerial image target detection technology can be used for practical applications in the military, traffic planning, personnel search and rescue and other fields. In this paper, we propose a multi-scale UAV aerial image detection method based on adaptive feature fusion for solving the problem of detecting small target objects in UAV aerial images. This method automatically adjusts the convolution kernel receptive field and reduces the redundant background of the image by adding an adaptive feature extraction module (AFEM) to the backbone network. This enables it to obtain more accurately and effectively small target feature information. In addition, we design an adaptive feature weighted fusion network (SBiFPN) to effectively enhance the representation of shallow feature information of small targets. Finally, we add an additional small target detection scale to the original network to expand the receptive field of the network and strengthen the detection of small target objects. The training and testing are carried out on the VisDrone public dataset. The experimental results show that the proposed method can achieve 38.5% mAP, which is 2.0% higher than the baseline network YOLOv5s, and can still detect the UAV aerial image well in complex scenes. [ABSTRACT FROM AUTHOR]

In this paper, a fault feature extraction method based on the combination of computational order tracking (COT) and variational mode decomposition (VMD) is proposed to solve the problem of fault feature extraction in a gearbox under variable speed conditions. First, a speed estimation method based on a forward-backward greedy algorithm and Fourier fitting is proposed to solve the speed estimation problem under variable speed conditions. Then, a feature extraction method based on COT, VMD, and fast kurtometry is proposed. COT is used to calculate the order. After the signal is decomposed by VMD, a fast kurtogram is used to filter the modal components with obvious fault features so as to extract fault features. Finally, combined with the analysis of actual equipment examples, the experimental results show that steady-state filtering can effectively filter the background noise and improve the signal-to-noise ratio. The rotation speed estimated by the forward–backward greedy algorithm and Fourier fitting is very close to the actual speed, which verifies that the method proposed in this paper can effectively solve the problem of fault feature extraction of a gearbox under variable speed conditions. [ABSTRACT FROM AUTHOR]

In order to solve the problem that the current image super‐resolution model has too many parameters and high computational complexity, this paper proposes a lightweight hybrid attention network (LHAN). LHAN consists of three parts: shallow feature extraction, lightweight hybrid attention block (LHAB), and upsampling module. LHAB combines multiscale self‐attention and large‐core attention. In order to make the network lightweight, multiscale self‐attention block (MSSAB) improves the self‐attention mechanism and uses windows of different sizes for group calculations. At the same time, in large‐core attention, we use depth‐based attention. Separate convolutions are used to reduce parameters. While keeping the receptive field unchanged, a normal convolution and a dilated convolution are used to replace the large kernel convolution. The four times super‐resolution experimental results on five data sets, including Set5 and Set14, show that our proposed method performs well in peak signal‐to‐noise ratio (PSNR) and structural similarity (SSIM). Specifically, in the benchmark data set on Urban, compared with SwinIR, the PSNR index of our method is improved by 0.10 dB. In addition, the parameter amount and calculation amount (floating point operations (FLOPs)) of our method are reduced by 315K and 16.4 G, respectively. Our proposed LHAN not only reduces the number of parameters and calculations but also achieves excellent performance in reconstruction quality. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Frontiers of Computer Science & Technology is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The abort mission refers to the mission where the landing vehicle needs to terminate the landing mission when an anomaly happens and be safely guided to the desired orbit. This paper focuses on solving the time-optimal abort guidance (TOAG) problem in real-time via the feature-based learning method. First, according to the optimal control theory, the features are identified to represent the optimal solutions of TOAG using a few parameters. After that, a sufficiently large data set of time-optimal abort trajectories is generated offline by solving the TOAG problems with different initial conditions. Then, the features are extracted for all generated cases. To find the implicit relationships between the initial conditions and identified features, neural networks are constructed to map the relationships based on the generated data set. Finally, experimental flight tests are conducted to demonstrate the onboard computation capability and effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Siamese-based trackers have been widely studied for their high accuracy and speed. Both the feature extraction and feature fusion are two important components in Siamese-based trackers. Siamese-based trackers obtain fine local features by traditional convolution. However, some important channel information and global information are lost when enhancing local features. In the feature fusion process, cross-correlation-based feature fusion between the template and search region feature ignores the global spatial context information and does not make the best of the spatial information. In this paper, to solve the above problem, we design a novel feature extraction sub-network based on batch-free normalization re-parameterization convolution, which scales the features in the channel dimension and increases the receptive field. Richer channel information is obtained and powerful target features are extracted for the feature fusion. Furthermore, we learn a feature fusion network (FFN) based on feature filter. The FFN fuses the template and search region features in a global spatial context to obtain high-quality fused features by enhancing important features and filtering redundant features. By jointly learning the proposed feature extraction sub-network and FFN, the local and global information are fully exploited. Then, we propose a novel tracking algorithm based on the designed feature extraction sub-network and FFN with re-parameterization convolution and feature filter, referred to as RCFT. We evaluate the proposed RCFT tracker and some recent state-of-the-art (SOTA) trackers on OTB100, VOT2018, LaSOT, GOT-10k, UAV123 and the visual-thermal dataset VOT-RGBT2019 datasets, which achieves superior tracking performance with 45 FPS tracking speed. [ABSTRACT FROM AUTHOR]

Scene-matching navigation is one of the essential technologies for achieving precise navigation in satellite-denied environments. Selecting suitable-matching areas is crucial for planning trajectory and reducing yaw. Most traditional selection methods of suitable-matching areas use hierarchical screening based on multiple feature indicators. However, these methods rarely consider the interrelationship between different feature indicators and use the same set of screening thresholds for different categories of images, which has poor versatility and can easily cause mis-selection and omission. To solve this problem, a suitable-matching areas' selection method based on multi-level saliency is proposed. The matching performance score is obtained by fusing several segmentation levels' salient feature extraction results and performing weighted calculations with the sub-image edge density. Compared with the hierarchical screening methods, the matching performance of the candidate areas selected by our algorithm is at least 22.2% higher, and it also has a better matching ability in different scene categories. In addition, the number of missed and wrong selections is significantly reduced. The average matching accuracy of the top three areas selected by our method reached 0.8549, 0.7993, and 0.7803, respectively, under the verification of multiple matching algorithms. Experimental results show this paper's suitable-matching areas' selection method is more robust. [ABSTRACT FROM AUTHOR]

Accurate recognition of tool state is important for maximizing tool life. However, the tool sensor data collected in real-life scenarios has unbalanced characteristics. Additionally, although graph neural networks (GNNs) show excellent performance in feature extraction in the spatial dimension of data, it is difficult to extract features in the temporal dimension efficiently. Therefore, we propose a tool state recognition method based on the Pruned Optimized Graph Neural Network-Gated Recurrent Unit (POGNN-GRU) under unbalanced data. Firstly, design the Improved-Majority Weighted Minority Oversampling Technique (IMWMOTE) by introducing an adaptive noise removal strategy and improving the MWMOTE to alleviate the unbalanced problem of data. Subsequently, propose a POG graph data construction method based on a multi-scale multi-metric basis and a Gaussian kernel weight function to solve the problem of one-sided description of graph data under a single metric basis. Then, construct the POGNN-GRU model to deeply mine the spatial and temporal features of the data to better identify the state of the tool. Finally, validation and ablation experiments on the PHM 2010 and HMoTP datasets show that the proposed method outperforms the other models in terms of identification, and the highest accuracy improves by 1.62% and 1.86% compared with the corresponding optimal baseline model. [ABSTRACT FROM AUTHOR]

As important disaster-bearing bodies, buildings are the focus of attention in seismic disaster risk assessment and emergency rescue. It is of great practical significance to extract buildings quickly and accurately with complex textures and variable scales and shapes from high-resolution remote sensing images. We proposed an improved TransUnet model based on multiscale grouped convolution and attention named MATUnet to retain more local detail features and enhance the representation ability of global features, while reducing the network parameters. We designed the multiscale grouped convolutional feature extraction module with attention (GAM) to enhance the representation of detailed features. The convolutional positional encoding module (PEG) was added to redetermine the number of transformer, it solved the problem of local feature information loss and the difficulty of convergence of the network. The channel attention module (CAM) of the decoder enhanced the salient information of the features and solved the problem of information redundancy after feature fusion. We experimented through MATUnet on the WHU building dataset and Massachusetts dataset. MATUnet achieved the best IOU results of 92.14% and 83.22%, respectively, and achieved better than the other generalized and state-of-the-art networks under the same conditions. We also have achieved good segmentation results on the GF2 Xichang building dataset. [ABSTRACT FROM AUTHOR]

Head Pose Estimation (HPE) has a wide range of applications in computer vision, but still faces challenges: (1) Existing studies commonly use Euler angles or quaternions as pose labels, which may lead to discontinuity problems. (2) HPE does not effectively address regression via rotated matrices. (3) There is a low recognition rate in complex scenes, high computational requirements, etc. This paper presents an improved unconstrained HPE model to address these challenges. First, a rotation matrix form is introduced to solve the problem of unclear rotation labels. Second, a continuous 6D rotation matrix representation is used for efficient and robust direct regression. The RepVGG-A2 lightweight framework is used for feature extraction, and by adding a multi-level feature fusion module and a coordinate attention mechanism with residual connection, to improve the network's ability to perceive contextual information and pay attention to features. The model's accuracy was further improved by replacing the network activation function and improving the loss function. Experiments on the BIWI dataset 7:3 dividing the training and test sets show that the average absolute error of HPE for the proposed network model is 2.41. Trained on the dataset 300W_LP and tested on the AFLW2000 and BIWI datasets, the average absolute errors of HPE of the proposed network model are 4.34 and 3.93. The experimental results demonstrate that the improved network has better HPE performance. [ABSTRACT FROM AUTHOR]

Non invasive load monitoring (NILM) is beneficial for enhancing the monitoring capability of the distribution network and is crucial for improving the safety of smart grid operation. However, household appliances involve a variety of devices and a large amount of data, making it difficult to achieve high-precision load identification. Especially when a large amount of loads are running simultaneously, problems such as feature overlap and reduced distinguish ability may occur, and thus increasing the difficulty of load identification. To solve these problems, this paper proposes a new NILM method based on dynamic time warping (DTW) optimization and event detection. Firstly, a feature extraction algorithm of STFT-SSAE is constructed by using short-time Fourier transform (STFT) to extract time–frequency features from the load, and then by sparse stack autoencoder (SSAE) to extract important features from time–frequency information. Secondly, the above features are input into Bi-LSTM and DTW models respectively, and a new probabilistic model is established. A Bi-LSTM-DTW load recognition architecture is built by combining the two models. Finally, the load identification model of SSAE-Bi-LSTM based on DTW optimization (DOSL) is trained by the preset combined data, which ensures the high confidence of the DOSL model in various complex operating scenarios. This paper introduces the public data set PLAID for experimental verification. The results show that the selected load features has good discrimination, and the proposed algorithm has the best identification effect with the accuracy rate of 0.9412. The proposed algorithm is also validated for generalization ability on the public dataset UK-DALE, with a identification accuracy rate of 0.9306, and the error of different datasets is only about 1%. • A complete multiple house-hold appliances NILM framework. • A novel STFT-SSAE algorithm for load feature extraction. • Bi-LSTM-DTW load identification architecture combined with output probability vector. [ABSTRACT FROM AUTHOR]

Image–text matching is a research hotspot in the multimodal task of integrating image and text processing. In order to solve the difficult problem of associating image and text data in the multimodal knowledge graph of Thangka, we propose an image and text matching method based on the Visual Semantic Embedding (VSE) model. The method introduces an adaptive pooling layer to improve the feature extraction capability of semantic associations between Thangka images and texts. We also improved the traditional Transformer architecture by combining bidirectional residual concatenation and mask attention mechanisms to improve the stability of the matching process and the ability to extract semantic information. In addition, we designed a multi-granularity tag alignment module that maps global and local features of images and text into a common coding space, leveraging inter- and intra-modal semantic associations to improve image and text accuracy. Comparative experiments on the Thangka dataset show that our method achieves significant improvements compared to the VSE baseline method. Specifically, our method improves the recall by 9.4% and 10.5% for image-matching text and text-matching images, respectively. Furthermore, without any large-scale corpus pre-training, our method outperforms all models without pre-training and outperforms two out of four pre-trained models on the Flickr30k public dataset. Also, the execution efficiency of our model is an order of magnitude higher than that of the pre-trained models, which highlights the superior performance and efficiency of our model in the image–text matching task. [ABSTRACT FROM AUTHOR]

• We proposed a small object detection layer (SODL), which focuses more on feature extraction for small objects. • We proposed a multi-scale receptive field (MSRF) module that extracts features around the nodules. • We proposed an efficient omni-dimensional convolution model, enabling attention weighting of the input data. (Aim) Aiming at the problem that there are a large number of small object nodules that are difficult to detect in lung images, detection methods based on improved YOLOv7 are proposed in this paper. (Method) First, a new small object detection layer (SODL) is proposed to solve the problem of the small size and irregular shape of lung nodules being difficult to detect accurately. Secondly, aiming at the problem that the characteristics of lung nodules are blurred and difficult to detect due to the continuous downsampling of the model, a multi-scale receptive field (MSRF) module is proposed and designed to improve the model's extraction of channel features. Finally, efficient omni-dimensional convolution (EODConv) is used to improve the ability of the network to extract the space, filters, and channels of the convolution kernel. (Results) Experiments were carried out on the public Luna16 dataset, and the results showed that our mAP, precision, and recall rate reached 95.26 %, 95.41 %, and 94.02 %, respectively, surpassing many state-of-the-art models. (Conclusion) In this study, a YOLOv7-based method is proposed for detecting lung nodules. Experimental results show that the proposed modification can significantly improve detection performance and is more suitable for clinical medical diagnosis. [ABSTRACT FROM AUTHOR]

Accurate identification of coastal hyperspectral remote sensing targets plays a significant role in the observation of marine ecosystems. Deep learning is currently widely used in hyperspectral recognition. However, most deep learning methods ignore the complex correlation and data loss problems that exist between features at different scales. In this study, Multi-scale attention reconstruction convolutional network (MARCN) is proposed to address the above issues. Firstly, a multi-scale attention mechanism is introduced into the network to optimize the feature extraction process, enabling the network to capture feature information at different scales and improve the target recognition performance. Secondly, the reconstruction module is introduced to fully utilize the spatial and spectral information of hyperspectral imagery, which effectively solves the problem of losing spatial and spectral information. Finally, an adaptive loss function, coupling cross-entropy loss, center loss, and feature space loss is used to enable the network to learn the feature representation and improve the accuracy of the model. The experimental results showed that the effectiveness of MARCN was validated with a recognition rate of 96.62%, and 97.92% on the YRE and GSOFF datasets. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy (DR), a leading cause of blindness in diabetic patients, necessitates the precise segmentation of lesions for the effective grading of lesions. DR multi-lesion segmentation faces the main concerns as follows. On the one hand, retinal lesions vary in location, shape, and size. On the other hand, the currently available multi-lesion region segmentation models are insufficient in their extraction of minute features and are prone to overlooking microaneurysms. To solve the above problems, we propose a novel deep learning method: the Multi-Scale Spatial Attention Gate (MSAG) mechanism network. The model inputs images of varying scales in order to extract a range of semantic information. Our innovative Spatial Attention Gate merges low-level spatial details with high-level semantic content, assigning hierarchical attention weights for accurate segmentation. The incorporation of the modified spatial attention gate in the inference stage enhances precision by combining prediction scales hierarchically, thereby improving segmentation accuracy without increasing the associated training costs. We conduct the experiments on the public datasets IDRiD and DDR, and the experimental results show that the proposed method achieves better performance than other methods. [ABSTRACT FROM AUTHOR]

Historical handwritten manuscripts pose challenges to automated recognition techniques due to their unique handwriting styles and cultural backgrounds. In order to solve the problems of complex text word misdetection, omission, and insufficient detection of wide-pitch curved text, this study proposes a high-precision text detection method based on improved YOLOv8s. Firstly, the Swin Transformer is used to replace C2f at the end of the backbone network to solve the shortcomings of fine-grained information loss and insufficient learning features in text word detection. Secondly, the Dysample (Dynamic Upsampling Operator) method is used to retain more detailed features of the target and overcome the shortcomings of information loss in traditional upsampling to realize the text detection task for dense targets. Then, the LSK (Large Selective Kernel) module is added to the detection head to dynamically adjust the feature extraction receptive field, which solves the cases of extreme aspect ratio words, unfocused small text, and complex shape text in text detection. Finally, in order to overcome the CIOU (Complete Intersection Over Union) loss in target box regression with unclear aspect ratio, insensitive to size change, and insufficient correlation between target coordinates, Gaussian Wasserstein Distance (GWD) is introduced to modify the regression loss to measure the similarity between the two bounding boxes in order to obtain high-quality bounding boxes. Compared with the State-of-the-Art methods, the proposed method achieves optimal performance in text detection, with the precision and mAP@0.5 reaching 86.3% and 82.4%, which are 8.1% and 6.7% higher than the original method, respectively. The advancement of each module is verified by ablation experiments. The experimental results show that the method proposed in this study can effectively realize complex text detection and provide a powerful technical means for historical manuscript reproduction. [ABSTRACT FROM AUTHOR]

Sentiment analysis is a fine‐grained analysis task that aims to identify the sentiment polarity of a specified sentence. Existing methods in Chinese sentiment analysis tasks only consider sentiment features from a single pole and scale and thus cannot fully exploit and utilise sentiment feature information, making their performance less than ideal. To resolve the problem, the authors propose a new method, GP‐FMLNet, that integrates both glyph and phonetic information and design a novel feature matrix learning process for phonetic features with which to model words that have the same pinyin information but different glyph information. Our method solves the problem of misspelling words influencing sentiment polarity prediction results. Specifically, the authors iteratively mine character, glyph, and pinyin features from the input comments sentences. Then, the authors use soft attention and matrix compound modules to model the phonetic features, which empowers their model to keep on zeroing in on the dynamic‐setting words in various positions and to dispense with the impacts of the deceptive‐setting ones. Experiments on six public datasets prove that the proposed model fully utilises the glyph and phonetic information and improves on the performance of existing Chinese sentiment analysis algorithms. [ABSTRACT FROM AUTHOR]

Most of the current tunnel defect detection methods do not perform repeated inspection of the defects taken at different times at the same location, which is a major deficiency for detecting the development trend of the defects. To solve the problem of repeated inspection of tunnel lining defects, this paper proposes a new method based on the combination of enhanced visual odometer and a Sequence-Siamese network(Seq-Siamese-Net) to identify and locate tunnel lining defects. We first detect the defects in the tunnel lining surface video by You Only Look Once network, then estimate the general location of the defects by using the proposed enhanced visual odometer, and finally obtain the most matching defects in the vicinity of the defects by Seq-Siamese-Net to achieve the repeated inspection of the tunnel lining surface defects. Experiments show that our proposed enhanced visual odometer method outperforms the original visual odometer method in terms of localization accuracy, while the feature extracted form Seq-Siamese-Network is more effective than the average-pooled feature in the tunnel defect image sequence matching task. Furthermore, experiments in a real tunnel demonstrate the practical value of the proposed method. • A network-based method for repeated detection of tunnel lining defects is proposed. • Average error of localization method reached about 1% at real tunnel experiment. • Sequence-Siamese-Network surpasses averaged feature in image sequence matching task. [ABSTRACT FROM AUTHOR]

Accurate detection and counting of flax plant organs are crucial for obtaining phenotypic data and are the cornerstone of flax variety selection and management strategies. In this study, a Flax-YOLOv5 model is proposed for obtaining flax plant phenotypic data. Based on the solid foundation of the original YOLOv5x feature extraction network, the network structure was extended to include the BiFormer module, which seamlessly integrates bi-directional encoders and converters, enabling it to focus on key features in an adaptive query manner. As a result, this improves the computational performance and efficiency of the model. In addition, we introduced the SIoU function to compute the regression loss, which effectively solves the problem of mismatch between predicted and actual frames. The flax plants grown in Lanzhou were collected to produce the training, validation, and test sets, and the detection results on the validation set showed that the average accuracy (mAP@0.5) was 99.29%. In the test set, the correlation coefficients (R) of the model's prediction results with the manually measured number of flax fruits, plant height, main stem length, and number of main stem divisions were 99.59%, 99.53%, 99.05%, and 92.82%, respectively. This study provides a stable and reliable method for the detection and quantification of flax phenotypic characteristics. It opens up a new technical way of selecting and breeding good varieties. [ABSTRACT FROM AUTHOR]

The existing Siamese trackers have achieved increasingly successful results in visual object tracking. However, the interactive fusion among multi-layer similarity maps after cross-correlation has not been fully studied in previous Siamese network-based methods. To address this issue, we propose a novel Siamese network for visual object tracking, named SiamSMN, which consists of a feature extraction network, a multi-scale fusion module, and a prediction head. First, the feature extraction network is used to extract the features of the template image and the search image, which is calculated by a depth-wise cross-correlation operation to produce multiple similarity feature maps. Second, we propose an effective multi-scale fusion module that can extract global context information for object search and learn the interdependencies between multi-level similarity maps. In addition, to further improve tracking accuracy, we design a learnable prediction head module to generate a boundary point for each side based on the coarse bounding box, which can solve the problem of inconsistent classification and regression during the tracking. Extensive experiments on four public benchmarks demonstrate that the proposed tracker has a competitive performance among other state-of-the-art trackers. [ABSTRACT FROM AUTHOR]

Sea–land segmentation (SLS) is a crucial step in coastline extraction. In CNN-based approaches for coastline feature extraction, downsampling is commonly used to reduce computational demands. However, this method may unintentionally discard small-scale features, hindering the capture of essential global contextual information and clear edge information necessary for SLS. To solve this problem, we propose a novel U-Net structure called Deformable Attention Edge Network (DAENet), which integrates edge enhancement algorithms and a deformable self-attention mechanism. First of all, we designed a multi-scale transformation (MST) to enhance edge feature extraction and model convergence through multi-scale transformation and edge detection, enabling the network to capture spatial–spectral changes more effectively. This is crucial because the deformability of the Deformable Attention Transformer (DAT) modules increases training costs for model convergence. Moreover, we introduced DAT, which leverages its powerful global modeling capabilities and deformability to enhance the model's recognition of irregular coastlines. Finally, we integrated the Local Adaptive Multi-Head Attention-based Edge Detection (LAMBA) module to enhance the spatial differentiation of edge features. We designed each module to address the complexity of SLS. Experiments on benchmark datasets demonstrate the superiority of the proposed DAENet over state-of-the-art methods. Additionally, we conducted ablation experiments to evaluate the effectiveness of each module. [ABSTRACT FROM AUTHOR]

Existing point cloud object detection algorithms struggle to effectively capture spatial features across different scales, often resulting in inadequate responses to changes in object size and limited feature extraction capabilities, thereby affecting detection accuracy. To solve this problem, we present a point cloud object detection method based on multi-scale feature fusion of the original point cloud and projection, which aims to improve the multi-scale performance and completeness of feature extraction in point cloud object detection. First, we designed a 3D feature extraction module based on the 3D Swin Transformer. This module pre-processes the point cloud using a 3D Patch Partition approach and employs a self-attention mechanism within a 3D sliding window, along with a downsampling strategy, to effectively extract features at different scales. At the same time, we convert the 3D point cloud to a 2D image using projection technology and extract 2D features using the Swin Transformer. A 2D/3D feature fusion module is then built to integrate 2D and 3D features at the channel level through point-by-point addition and vector concatenation to improve feature completeness. Finally, the integrated feature maps are fed into the detection head to facilitate efficient object detection. Experimental results show that our method has improved the average precision of vehicle detection by 1.01% on the KITTI dataset over three levels of difficulty compared to Voxel-RCNN. In addition, visualization analyses show that our proposed algorithm also exhibits superior performance in object detection. [ABSTRACT FROM AUTHOR]

Weed detection is closely related to agricultural production, but often faces the problems of leaf shading and limited computational resources. Therefore, this study proposes an improved weed detection algorithm based on YOLOv8. Firstly, the Dilated Feature Integration Block is designed to improve the feature extraction in the backbone network by introducing large kernel convolution and multi-scale dilation convolution, which utilizes information from different scales and levels. Secondly, to solve the problem of a large number of parameters in the feature fusion process of the Path Aggregation Feature Pyramid Network, a new feature fusion architecture multi-scale feature interaction network is designed, which achieves the high-level semantic information to guide the low-level semantic information through the attention mechanism. Finally, we propose a Dynamic Feature Aggregation Head to solve the problem that the YOLOv8 detection head cannot dynamically focus on important features. Comprehensive experiments on two publicly accessible datasets show that the proposed model outperforms the benchmark model, with mAP50 and mAP75 improving by 4.7% and 5.0%, and 5.3% and 3.3%, respectively, whereas the number of model parameters is only 6.62 M. This study illustrates the utility potential of the algorithm for weed detection in cotton fields, marking a significant advancement of artificial intelligence in agriculture. [ABSTRACT FROM AUTHOR]

Remote sensing image change detection (CD) is an important means in remote sensing data analysis tasks, which can help us understand the surface changes in high-resolution (HR) remote sensing images. Traditional pixel-based and object-based methods are only suitable for low- and medium-resolution images, and are still challenging for complex texture features and detailed image detail processing in HR images. At present, the method based on deep learning has problems such as inconsistent fusion and difficult model training in the combination of the difference feature information of the deep and shallow layers and the attention mechanism, which leads to errors in the distinction between the changing region and the invariant region, edge detection and small target detection. In order to solve the above problems of inconsistent fusions of feature information aggregation and attention mechanisms, and indistinguishable change areas, we propose a multi-scale feature fusion Siamese network based on attention mechanism (ABMFNet). To tackle the issues of inconsistent fusion and alignment difficulties when integrating multi-scale fusion and attention mechanisms, we introduce the attention-based multi-scale feature fusion module (AMFFM). This module not only addresses insufficient feature fusion and connection between different-scale feature layers, but also enables the model to automatically learn and prioritize important features or regions in the image. Additionally, we design the cross-scale fusion module (CFM) and the difference feature enhancement pyramid structure (DEFPN) to assist the AMFFM module in integrating differential information effectively. These modules bridge the spatial disparity between low-level and high-level features, ensuring efficient connection and fusion of spatial difference information. Furthermore, we enhance the representation and inference speed of the feature pyramid by incorporating a feature enhancement module (FEM) into DEFPN. Finally, the BICD dataset proposed by the laboratory and public datasets LEVIR-CD and BCDD are compared and tested. We use F1 score and MIoU values as evaluation metrics. For AMBMFNet, the F1 scores on the three datasets are 77.69%, 81.57%, and 77.91%, respectively, while the MIoU values are 84.65%, 85.84%, and 84.54%, respectively. The experimental results show that ABMFNet has better effectiveness and robustness. [ABSTRACT FROM AUTHOR]

Accurate and robust localization is one of the most basic functional modules of mobile robots. A recent promising approach is to employ a camera that captures ground textures to achieve high-precision localization. However, low robustness and real-time performance are disadvantages of the system due to the time-consuming front-end feature extraction and search methods. To solve these problems, a comprehensive system that has high robustness and real-time performance is proposed, including offline mapping, global localization, local localization and map update. This paper summarizes a method for selecting the optimal front-end for different textures. A fast global localization method based on bag of words is proposed, which can quickly and effectively determine the local positioning area. Experiments are carried out on public datasets and actual indoor scenes to verify the practicability of the system. [ABSTRACT FROM AUTHOR]

During recent years, deep convolutional neural networks (CNNs) have become increasingly popular in sensor-based human activity recognition (HAR), due to powerful feature extraction capacity. However, the required high computational cost has hindered their practical deployment in real-world applications. To address this issue, this paper presents a novel approach named ABCSearch to prune channels through an artificial bee colony (ABC) algorithm, which is a meta-heuristic optimization algorithm inspired by the foraging behavior of honeybees. Unlike traditional pruning strategies that prioritize channel importance, the proposed method seeks to identify the optimal number of channels at each layer, i.e., the optimal pruning structure. To lessen human interference, we transform the search problem of the optimal pruning structure into a simple optimization problem and solve it automatically through the ABC algorithm, which can significantly decrease both model size and computational cost without compromising performance. Specifically, we validate the effectiveness of ABCSearch for a standard five-layer CNN backbone architecture, which achieves 97.05%, 98.45%, 91.71%, 78.89% classification accuracies on four HAR benchmarks, i.e., UCI-HAR, WISDM, PAMAP2, UniMiB-SHAR. Our approach can yield 87.44%, 89.41%, 89.06%, 89.59% FLOPs reductions, and 89.54%, 88.70%, 88.38%, 90.45% less memory consumptions, as well as 4.4 × , 5.0 × , 5.8 × , 6.1 × practical speedups on real hardware devices, while maintaining comparable accuracy. [ABSTRACT FROM AUTHOR]

Accurate prediction of formation pore pressure is essential to predict fluid flow and manage hydrocarbon production in petroleum engineering. Recent deep learning technique has been receiving more interest due to the great potential to deal with pore pressure prediction. However, most of the traditional deep learning models are less efficient to address generalization problems. To fill this technical gap, in this work, we developed a new adaptive physics-informed deep learning model with high generalization capability to predict pore pressure values directly from seismic data. Specifically, the new model, named CGP-NN, consists of a novel parametric features extraction approach (1DCPP), a stacked multilayer gated recurrent model (multilayer GRU), and an adaptive physics-informed loss function. Through machine training, the developed model can automatically select the optimal physical model to constrain the results for each pore pressure prediction. The CGP-NN model has the best generalization when the physicsrelated metric λ = 0:5. A hybrid approach combining Eaton and Bowers methods is also proposed to build machine-learnable labels for solving the problem of few labels. To validate the developed model and methodology, a case study on a complex reservoir in Tarim Basin was further performed to demonstrate the high accuracy on the pore pressure prediction of new wells along with the strong generalization ability. The adaptive physics-informed deep learning approach presented here has potential application in the prediction of pore pressures coupled with multiple genesis mechanisms using seismic data. [ABSTRACT FROM AUTHOR]

Recently, owing to the requirements of inference speed, most real-time semantic segmentation networks often have shallow network depth, which limits the receptive field size of the model, leading to the limited acquisition of semantic information and resulting in intraclass inconsistency and ultimately a decrease in segmentation accuracy. Additionally, the shallow network depth also restricts the feature extraction capability of the network, reducing its robustness and ability to adapt to complex scenes. To address these issues, a bilateral network with a rich semantic extractor (RSE) for real-time semantic segmentation (BRSeNet) is presented to perform real-time semantic segmentation. First, to solve the problem of insufficient semantic feature information extraction, an RSE is proposed, which includes a multiscale global semantic extraction module (MGSEM) and a semantic fusion module (SFM). The MGSEM can extract rich global semantics and expand the effective receptive field. Simultaneously, the SFM efficiently integrates multiscale local semantics with multiscale global semantics, resulting in more comprehensive semantic information for the network. Finally, based on the characteristics of detail and semantic branches, a bilateral reconstruction aggregation module is designed to reconstruct the contextual information of detail features, model the interdependencies on semantic feature channels, and enhance feature representation. Comprehensive experiments on the challenging Cityscapes and ADE20K datasets are conducted. The experimental results show that the proposed BRSeNet achieves mean intersection over union of 74.9% and 35.7% at inference speeds of 74 and 65 frames per second, respectively, and ensures a favorable balance between segmentation accuracy and inference speed. [ABSTRACT FROM AUTHOR]

Chinese is characterized by high syntactic complexity, chaotic annotation granularity, and slow convergence. Joint learning models can effectively improve the accuracy of Chinese Named Entity Recognition (NER), but they focus too much on local feature information and reduce the ability of long sequence feature extraction. To address the limitations of long sequence feature extraction ability, we propose a Chinese NER model called Incorporating Recurrent Cell and Information State Recursion (IRCSR-NER). The model integrates recurrent cells and information state recursion to improve the recognition ability of long entity boundaries. To solve the problem that Chinese and English have different focuses in syntactic analysis. We use the syntactic dependency approach to add lexical relationship information to sentences represented at the word level. The IRCSR-NER is applied to sequence feature extraction to improve the model efficiency and long-text feature extraction ability. The model captures contextual long-distance dependent information while focusing on local feature information. We evaluated our proposed model using four public datasets and compared it with other mainstream models. Experimental results demonstrate that our model outperforms traditional and mainstream models. [ABSTRACT FROM AUTHOR]

It is very challenging to accurately segment smoke images because smoke has some adverse vision characteristics, such as anomalous shapes, blurred edges, and translucency. Existing methods cannot fully focus on the texture details of anomalous shapes and blurred edges simultaneously. To solve these problems, a Dense Multi‐scale context and Asymmetric pooling Embedding Network (DMAENet) is proposed to model the smoke edge details and anomalous shapes for smoke segmentation. To capture the feature information from different scales, a Dense Multi‐scale Context Module (DMCM) is proposed to further enhance the feature representation capability of our network under the help of asymmetric convolutions. To efficiently extract features for long‐shaped objects, the authors use asymmetric pooling to propose an Asymmetric Pooling Enhancement Module (APEM). The vertical and horizontal pooling methods are responsible for enhancing features of irregular objects. Finally, a Feature Fusion Module (FFM) is designed, which accepts three inputs for improving performance. Low and high‐level features are fused by pixel‐wise summing, and then the summed feature maps are further enhanced in an attention manner. Experimental results on synthetic and real smoke datasets validate that all these modules can improve performance, and the proposed DMAENet obviously outperforms existing state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

Accurate identification of apple leaf diseases is of great significance for improving apple yield. The lesion area of the apple leaf disease image is small and vulnerable to background interference, which easily leads to low recognition accuracy. To solve this problem, a lightweight bilinear convolutional neural network (CNN) model named BLSENet based on attention mechanism is designed. The model consists of two subnetworks, and each subnetwork is embedded with a Squeeze-and-Excitation (SE) module. By using the feature extraction ability of the two subnetworks and combining the bilinear feature CONCAT operation, the multiscale features of the image are obtained. Compared with the unimproved model LeNet-5 (84.63%), BLSENet has higher accuracy in the test set, which indicates that SE module and bilinear feature fusion have a positive effect on the performance of the model, and BLSENet has the ability to identify apple leaf diseases. The model has achieved the expected goal and can provide technical support for accurate identification and real-time monitoring of apple disease images. [ABSTRACT FROM AUTHOR]

Joint entity and relation extraction methods have attracted an increasing amount of attention recently due to their capacity to extract relational triples from intricate texts. However, most of the existing methods ignore the association and difference between the Named Entity Recognition (NER) subtask features and the Relation Extraction (RE) subtask features, which leads to an imbalance in the interaction between these two subtasks. To solve the above problems, we propose a new joint entity and relation extraction method, FSN. It contains a Filter Separator Network (FSN) module that employs a two-direction LSTM to filter and separate the information contained in a sentence and merges similar features through a splicing operation, thus solving the problem of the interaction imbalance between subtasks. In order to better extract the local feature information for each subtask, we designed a Named Entity Recognition Generation (NERG) module and a Relation Extraction Generation (REG) module by adopting the design idea of the decoder in Transformer and average pooling operations to better capture the entity boundary information in the sentence and the entity pair boundary information for each relation in the relational triple, respectively. Additionally, we propose a dynamic loss function that dynamically adjusts the learning weights of each subtask in each epoch according to the proportionality between each subtask, thus narrowing down the difference between the ideal and realistic results. We thoroughly evaluated our model on the SciERC dataset and the ACE2005 dataset. The experimental results demonstrate that our model achieves satisfactory results compared to the baseline model. [ABSTRACT FROM AUTHOR]

In the process of the individual identification of radiation sources, the effective extraction of fine features of target radiation sources can be regarded as being crucial for the subsequent individual identification. However, in the complex electromagnetic environment, the effective extraction of the radiation source features is still facing great challenges. To solve this problem, we propose an algorithm for constructing the attractor feature space of the radiation source system based on blind equalization to solve this problem. Firstly, we use blind equalization to process the target signal. Secondly, we use the phase-space reconstruction technique to construct the system attractor feature space of the target signal processed, and explore the adaptive relationship between feature-space-embedding dimensions, the delay time and the neural network, finding the optimal values. Finally, the feature space is used as the input of the neural network for the subsequent individual discrimination decision. Experimentally, it is proved that our proposed algorithm improves the individual recognition rate of the target radiation source under the complex electromagnetic environment to a certain extent, which is of practical application value. [ABSTRACT FROM AUTHOR]

Deep learning models have been widely applied in the field of crop disease recognition. There are various types of crops and diseases, each potentially possessing distinct and effective features. This brings a great challenge to the generalization performance of recognition models and makes it very difficult to build a unified model capable of achieving optimal recognition performance on all kinds of crops and diseases. In order to solve this problem, we have proposed a novel ensemble learning method for crop leaf disease recognition (named ELCDR). Unlike the traditional voting strategy of ensemble learning, ELCDR assigns different weights to the models based on their feature extraction performance during ensemble learning. In ELCDR, the models' feature extraction performance is measured by the distribution of the feature vectors of the training set. If a model could distinguish more feature differences between different categories, then it receives a higher weight during ensemble learning. We conducted experiments on the disease images of four kinds of crops. The experimental results show that in comparison to the optimal single model recognition method, ELCDR improves by as much as 1.5 (apple), 0.88 (corn), 2.25 (grape), and 1.5 (rice) percentage points in accuracy. Compared with the voting strategy of ensemble learning, ELCDR improves by as much as 1.75 (apple), 1.25 (corn), 0.75 (grape), and 7 (rice) percentage points in accuracy in each case. Additionally, ELCDR also has improvements on precision, recall, and F1 measure metrics. These experiments provide evidence of the effectiveness of ELCDR in the realm of crop leaf disease recognition. [ABSTRACT FROM AUTHOR]

The maintenance level of rural roads is relatively low, and the automated detection of pavement distress is easily affected by the shadows of rows of trees, weeds, soil, and distress object scale disparities; this makes it difficult to accurately evaluate the distress conditions of the pavement. To solve the above problems, this study specifically designed a target detection network called Crack Convolution (CrackYOLO) for pavement crack extraction on rural roads. CrackYOLO is based on an improved YOLOv5. The shadow created by rows of trees leads to the loss of crack features in the feature extraction and downsampling stages of the network; therefore, CrackConv and Adapt-weight Down Sample (ADSample) were introduced to strengthen the ability to locate and identify cracks. Due to disturbances such as soil and weeds, which cause the extraction of more redundant features, the Channel And Spatial mixed attention mechanism (CAS) was introduced to enhance crack weight. To address the issue of missed detections of fine cracks due to significant scale variations in crack objects in the same image, Multi Scale Convolution (MSConv) and Multi Scale Head (MSHead) were incorporated during the feature fusion and prediction inference stages of the network, thereby improving the multi-scale detection performance. In order to verify the effectiveness of the proposed method, the detection accuracy of CrackYOLO when used on the LNTU_RDD_NC dataset was determined to be 9.99%, 12.79%, and 4.61% higher than that of the current pavement crack detection models YOLO-LWNet, Faster R-CNN, and YOLOv7. At the same time, we compare the above model on public datasets of different scenarios, and the experimental results show that CrackYOLO has the same strong performance in urban roads and other scenarios. [ABSTRACT FROM AUTHOR]

Gait recognition has received widespread attention due to its non-intrusive recognition mechanism. Currently, most gait recognition methods use appearance-based recognition methods, and such methods are easily affected by occlusions when facing complex environments, which in turn affects the recognition accuracy. With the maturity of pose estimation techniques, model-based gait recognition methods have received more and more attention due to their robustness in complex environments. However, the current model-based gait recognition methods mainly focus on modeling the global feature information in the spatial dimension, ignoring the importance of local features and their influence on recognition accuracy. Meanwhile, in the temporal dimension, these methods usually use single-scale temporal information extraction, which does not take into account the inconsistency of the motion cycles of the limbs when a human body is walking (e.g., arm swing and leg pace), leading to the loss of some limb temporal information. To solve these problems, we propose a gait recognition network based on a Global–Local Graph Convolutional Network, called GaitMGL. Specifically, we introduce a new spatio-temporal feature extraction module, MGL (Multi-scale Temporal and Global–Local Spatial Extraction Module), which consists of GLGCN (Global–Local Graph Convolutional Network) and MTCN (Multi-scale Temporal Convolutional Network). GLGCN models both global and local features, and extracts global–local motion information. MTCN, on the other hand, takes into account the inconsistency of local limb motion cycles, and facilitates multi-scale temporal convolution to capture the temporal information of limb motion. In short, our GaitMGL solves the problems of loss of local information and loss of temporal information at a single scale that exist in existing model-based gait recognition networks. We evaluated our method on three publicly available datasets, CASIA-B, Gait3D, and GREW, and the experimental results show that our method demonstrates surprising performance and achieves an accuracy of 63.12% in the dataset GREW, exceeding all existing model-based gait recognition networks. [ABSTRACT FROM AUTHOR]

The popularity of Agrocybe cylindracea is increasing due to its unique flavor and nutritional value. The Agrocybe cylindracea cap is a key aspect of the growth process, and high-throughput observation of cap traits in greenhouses by machine vision is a future development trend of smart agriculture. Nevertheless, the segmentation of the Agrocybe cylindracea cap is extremely challenging due to its similarity in color to the rest of the mushroom and the occurrence of mutual occlusion, presenting a major obstacle for the effective application of automation technology. To address this issue, we propose an improved instance segmentation network called Agrocybe cylindracea R-CNN (AC R-CNN) based on the Mask R-CNN model. AC R-CNN incorporates hybrid dilated convolution (HDC) and attention modules into the feature extraction backbone network to enhance the segmentation of adhesive mushroom caps and focus on the segmentation objects. Furthermore, the Mask Branch module is replaced with PointRend to improve the network's segmentation accuracy at the edges of the mushroom caps. These modifications effectively solve the problems of the original algorithm's inability to segment adhesive Agrocybe cylindracea caps and low accuracy in edge segmentation. The experimental results demonstrate that AC R-CNN outperforms the original Mask R-CNN in terms of segmentation performance. The average precision (AP) is improved by 12.1%, and the F1 score is improved by 13.7%. Additionally, AC R-CNN outperforms other networks such as Mask Scoring R-CNN and BlendMask. Therefore, the research findings of this study can meet the high-precision segmentation requirements of Agrocybe cylindracea caps and lay a theoretical foundation for the development of subsequent intelligent phenotyping devices and harvesting robots. [ABSTRACT FROM AUTHOR]

Traditional human pose estimation methods typically rely on complex models and algorithms. Lite-HRNet can achieve an excellent performance while reducing model complexity. However, its feature extraction scale is relatively single, which can lead to lower keypoints' localization accuracy in crowded and complex scenes. To address this issue, we propose a lightweight human pose estimation model based on a joint channel coordinate attention mechanism. This model provides a powerful information interaction channel, enabling features of different resolutions to interact more effectively. This interaction can solve the problem of human pose estimation in complex scenes and improve the robustness and accuracy of the pose estimation model. The introduction of the joint channel coordinate attention mechanism enables the model to more effectively retain key information, thereby enhancing keypoints' localization accuracy. We also redesign the lightweight basic module using the shuffle module and the joint channel coordinate attention mechanism to replace the spatial weight calculation module in the original Lite-HRNet model. By introducing this new module, we not only improve the network calculation speed and reduce the number of parameters of the entire model, but also ensure the accuracy of the model, thereby achieving a balance between performance and efficiency. We compare this model with current mainstream methods on the COCO and MPII dataset. The experimental results show that this model can effectively reduce the number of parameters and computational complexity while ensuring high model accuracy. [ABSTRACT FROM AUTHOR]