Extract Informative Features Research Articles

SSL-VQ: Vector-quantized variational autoencoders for semi-supervised prediction of therapeutic targets across diverse diseases.

Identifying effective therapeutic targets poses a challenge in drug discovery, especially for uncharacterized diseases without known therapeutic targets (e.g. rare diseases, intractable diseases). This study presents a novel machine learning approach employing multimodal vector-quantized variational autoencoders (VQ-VAEs) for predicting therapeutic target molecules across diseases. To address the lack of known therapeutic target-disease associations, we incorporate the information on uncharacterized diseases without known targets or uncharacterized proteins without known indications (applicable diseases) in the semi-supervised learning (SSL) framework. The method integrates disease-specific and protein perturbation profiles with genetic perturbations (e.g., gene knockdowns and gene overexpressions) at the transcriptome level. Cross-cell representation learning, facilitated by VQ-VAEs, was performed to extract informative features from protein perturbation profiles across diverse human cell types. Concurrently, cross-disease representation learning was performed, leveraging VQ-VAE, to extract informative features reflecting disease states from disease-specific profiles. The model's applicability to uncharacterized diseases or proteins is enhanced by considering consistency between disease-specific and patient-specific signatures. The efficacy of the method is demonstrated across three practical scenarios for 79 diseases: target repositioning for target-disease pairs, new target prediction for uncharacterized diseases, and new indication prediction for uncharacterized proteins. This method is expected to be valuable for identifying therapeutic targets across various diseases. Code: github.com/YamanishiLab/SSL-VQ & Data: 10.5281/zenodo.14644837. Supplementary data are available at Bioinformatics online.

ARCH: Large-scale knowledge graph via aggregated narrative codified health records analysis.

Electronic health record (EHR) systems contain a wealth of clinical data stored as both codified data and free-text narrative notes (NLP). The complexity of EHR presents challenges in feature representation, information extraction, and uncertainty quantification. To address these challenges, we proposed an efficient Aggregated naRrative Codified Health (ARCH) records analysis to generate a large-scale knowledge graph (KG) for a comprehensive set of EHR codified and narrative features. Using data from 12.5 million Veterans Affairs patients, ARCH first derives embedding vectors and generates similarities along with associated p-values to measure the strength of relatedness between clinical features with statistical certainty quantification. Next, ARCH performs a sparse embedding regression to remove indirect linkage between features to build a sparse KG. Finally, ARCH was validated on various clinical tasks, including detecting known relationships between entity pairs, predicting drug side effects, disease phenotyping, as well as sub-typing Alzheimer's disease patients. ARCH produces high-quality clinical embeddings and KG for over 60,000 codified and narrative EHR concepts. The KG and embeddings are visualized in the R-shiny powered web-API3. ARCH achieved high accuracy in detecting EHR concept relationships, with AUCs of 0.926 (codified) and 0.861 (NLP) for similar EHR concepts, and 0.810 (codified) and 0.843 (NLP) for related pairs. It detected drug side effects with a 0.723 AUC, which improved to 0.826 after fine-tuning. Using both codified and NLP features, the detection power increased significantly. Compared to other methods, ARCH has superior accuracy and enhances weakly supervised phenotyping algorithms' performance. Notably, it successfully categorized Alzheimer's patients into two subgroups with varying mortality rates. The proposed ARCH algorithm generates large-scale high-quality semantic representations and knowledge graph for both codified and NLP EHR features, useful for a wide range of predictive modeling tasks.

Intelligent fault diagnosis based on multi-sensor data fusion and multi-scale dual attention enhanced network

Abstract As a significant component of rotating machinery, the health of rolling bearings and gears directly impact the normal operation of rotating machinery. Traditional single-sensor fault diagnosis approaches often fail to extract sufficient fault information, resulting in low diagnostic accuracy in practical engineering applications. Additionally, conventional multi-sensor fusion diagnosis methods exhibit the low robustness in noisy environments. To tackle these challenges, this paper presents a novel fault diagnosis approach derived from multi-sensor data fusion and a multi-scale dual attention enhanced network (MSDF-MSDAENet). Initially, a multi-sensor data fusion (MSDF) strategy is presented. This method reduces the dimensionality of high-dimensional data from multiple sensors to extract embedded low-dimensional effective information, which is then fused into a three-dimensional pixel matrix image. Subsequently, a multi-scale dual attention enhanced network (MSDAENet) is constructed, incorporating both time-frequency information feature extraction modules and position information extraction modules. These enhancements significantly enhance the model's capacity to extract fault features. By dividing the RGB fused images into a well-organized dataset and inputting it into the network for fault diagnosis, the method achieves intelligent diagnosis of samples with different fault types. Tests were carried out on three datasets, and the presented method was compared with various existing methods. The ultimate experimental results demonstrate the efficacy and superiority of the presented method.

Multi-Scale Building Load Forecasting Without Relying on Weather Forecast Data: A Temporal Convolutional Network, Long Short-Term Memory Network, and Self-Attention Mechanism Approach

Accurate load forecasting is of vital importance for improving the energy utilization efficiency and economic profitability of intelligent buildings. However, load forecasting is restricted in the popularization and application of conventional load forecasting techniques due to the great difficulty in obtaining numerical weather prediction data at the hourly level and the requirement to conduct predictions on multiple time scales. Under the condition of lacking meteorological forecast data, this paper proposes to utilize a temporal convolutional network (TCN) to extract the coupled spatial features among multivariate loads. The reconstructed features are then input into the long short-term memory (LSTM) neural network to achieve the extraction of load time features. Subsequently, the self-attention mechanism is employed to strengthen the model’s ability to extract feature information. Finally, load forecasting is carried out through a fully connected network, and a multi-time scale prediction model for building multivariate loads based on TCN–LSTM–self-attention is constructed. Taking a hospital building as an example, this paper predicts the cooling, heating, and electrical loads of the hospital for the next 1 h, 1 day, and 1 week. The experimental results show that on multiple time scales, the TCN–LSTM–self-attention prediction model proposed in this paper is more accurate than the LSTM, CNN-LSTM, and TCN-LSTM models. Especially in the task of predicting cooling, heating, and electrical loads on a 1-week scale, the model proposed in this paper achieves improvements of 16.58%, 6.77%, and 3.87%, respectively, in the RMSE indicator compared with the TCN-LSTM model.

CFNet:CSI compression feedback network based on WiFi sensing

Abstract To solve the problem of huge channel state information (CSI)data caused by a large number of antennas in large-scale MIMO systems,a CSI compressed feedback network model (CFNet)with encoder-decoder structure is proposed in this paper. Specifically, the encoder efficiently compresses the CSI data to reduce the burden of data transmission and storage. A key component, the Asymmetric Expansion Convolution (AEC) module, combines asymmetric convolution and dilated convolution to extract feature information from different directions, thereby increasing the model's receptive field and minimizing information loss. The compressed CSI is fed into a Gated Recurrent Unit (GRU), which captures long-term dependencies in the CSI sequence using the encoder's output and the hidden state of the upper window. Additionally, on the decoder side, the proposed Asymmetric Transpose Convolution reconstructs the compressed CSI data, enhancing the network's ability to capture and express CSI features through two complementary convolution operations, ultimately reducing communication costs. Thanks to the proposed encoder and decoder, CFNet can accurately identify fine-grained features of human activities recognition(HAR) while leveraging compressed sensing techniques, further improving data processing efficiency and recognition accuracy. The experiments conducted on two public datasets demonstrate that the evaluation metrics of the proposed network outperform those of advanced methods in the relevant field.

DualTransAttNet: A Hybrid Model with a Dual Attention Mechanism for Corn Seed Classification

Varietal purity is a critical quality indicator for seeds, yet various production processes can lead to the mixing of seeds from different varieties. Consequently, seed variety classification is an essential step in seed production. Existing classification algorithms often suffer from limitations such as reliance on single information sources, constrained feature extraction capabilities, time consumption, low accuracy, and the potential to cause irreversible damage to seeds. To address these challenges, this paper proposes a fast and non-destructive classification method for corn seeds, named DualTransAttNet, based on multi-source image information and hybrid feature extraction. High-resolution hyperspectral images of various corn varieties were collected, and a sliding sampling approach was employed to capture feature information across all spectral bands, resulting in the construction of a hyperspectral dataset for corn seed classification. Hyperspectral and RGB image data were then integrated to complement one another’s information and mitigate the insufficient feature diversity caused by single-source data. The proposed method leverages the strengths of convolutional neural networks (CNNs) and transformers to extract both local and global features, effectively capturing spectral and image characteristics. The experimental results demonstrate that the DualTransAttNet model can achieve a compact size of only 1.758 MB and an inference time of 0.019 ms. Compared to typical machine learning and deep learning models, the proposed model exhibits superior performance with an overall accuracy, F1-score, and Kappa coefficient of 90.01%, 88.9%, and 88.4%, respectively. The model’s rapid inference capability and low parameter count make it an excellent technical solution for agricultural automation and intelligent systems, thereby enhancing the efficiency and profitability of agricultural production.

Wind Power Prediction Based on The SENet-CNN-LSTM Model

aims: To improve the accuracy of wind power prediction background: In recent years, wind power has developed rapidly nationwide and worldwide.How to improve the prediction precision and accuracy of wind power has become a research hotspot in this field objective: This paper aims to construct a wind power forecasting system based on SENet-CNN-LSTM to improve the accuracy and stability of wind power forecasting method: Firstly, the isolated forest algorithm is used to detect abnormal values in wind power historical data, and the linear interpolation method is used to fill in the missing data. Secondly, CNN is used to extract the spatio-temporal characteristics of wind power data, SENet is used to assign different weights to the extracted feature information, and LSTM’s unique gating mechanism is used to memorize and forget the information. Finally, taking the measured historical data of a wind power farm as the sample, six algorithms including CNN, LSTM, CNN-LSTM, SENet-CNN, SENet-LSTM and SENet-CNN-LSTM are used to predict the example result: The prediction results show that the RMSE, MAE and MAPE of SENet-CNN-LSTM are significantly reduced compared with the other five prediction models conclusion: The wind power prediction model based on SENet-CNN-LSTM has high prediction accuracy and good performance other: There is no.

Advanced analytical methods for multi-spectral transmission imaging optimization: enhancing breast tissue heterogeneity detection and tumor screening with hybrid image processing and deep learning.

Light sources exhibit significant absorption and scattering effects during the transmission through biological tissues, posing challenges in identifying heterogeneities in multi-spectral images. This paper introduces a fusion of techniques encompassing the spatial pyramid matching model (SPM), modulation and demodulation (M_D), and frame accumulation (FA). These techniques not only elevate image quality but also augment the precision of heterogeneous classification in multi-spectral transmission images (MTI) within deep learning network models (DLNM). Initially, experiments are designed to capture MTI of phantoms. Subsequently, the images are preprocessed separately through a combination of different techniques such as SPM, M_D and FA. Ultimately, multi-spectral fusion pseudo-color images derived from U-Net semantic segmentation are fed into VGG16/19 and ResNet50/101 networks for heterogeneous classification. Among them, different combinations of SPM, M_D and FA significantly enhance the quality of images, facilitating the extraction of heterogeneous feature information from multi-spectral images. In comparison to the classification accuracy achieved in the original image VGG and ResNet network models, all images after preprocessing effectively improved the classification accuracy of heterogeneities. Following scatter correction, images processed with 3.5 Hz modulation-demodulation combined with frame accumulation (M_D-FA) attain the highest classification accuracy for heterogeneities in the VGG19 and ResNet101 models, achieving accuracies of 95.47% and 98.47%, respectively. In conclusion, this paper utilizes different combinations of SPM, M_D and FA techniques to not only enhance the quality of images but also further improve the accuracy of DLNM in heterogeneous classification, which will promote the clinical application of MTI technique in breast tumor screening.

Multiscale Residual Weighted Classification Network for Human Activity Recognition in Microwave Radar.

Human activity recognition by radar sensors plays an important role in healthcare and smart homes. However, labeling a large number of radar datasets is difficult and time-consuming, and it is difficult for models trained on insufficient labeled data to obtain exact classification results. In this paper, we propose a multiscale residual weighted classification network with large-scale, medium-scale, and small-scale residual networks. Firstly, an MRW image encoder is used to extract salient feature representations from all time-Doppler images through contrastive learning. This can extract the representative vector of each image and also obtain the pre-training parameters of the MRW image encoder. During the pre-training process, large-scale residual networks, medium-scale residual networks, and small-scale residual networks are used to extract global information, texture information, and semantic information, respectively. Moreover, the time-channel weighting mechanism can allocate weights to important time and channel dimensions to achieve more effective extraction of feature information. The model parameters obtained from pre-training are frozen, and the classifier is added to the backend. Finally, the classifier is fine-tuned using a small amount of labeled data. In addition, we constructed a new dataset with eight dangerous activities. The proposed MRW-CN model was trained on this dataset and achieved a classification accuracy of 96.9%. We demonstrated that our method achieves state-of-the-art performance. The ablation analysis also demonstrated the role of multi-scale convolutional kernels and time-channel weighting mechanisms in classification.

Lifespan prediction model for proton exchange membrane fuel cell vehicle based on time series information feature extraction and optimization

Lifespan prediction model for proton exchange membrane fuel cell vehicle based on time series information feature extraction and optimization

Development of a universal multimodal prediction method to optimise process parameters for improving densification during laser powder bed fusion

ABSTRACT This work develops a universal machine learning method that enhances the densification of metallic materials fabricated by laser powder bed fusion (LPBF) by optimising process parameters using multimodal data. By integrating ResNet-50 neural network and LightGBM model, this method combines process parameters with surface morphology of LPBF-fabricated samples at the feature scale, enabling extraction of multidimensional feature information and improving prediction accuracy of relative density. Statistical analysis demonstrates exceptional performance of the model on both training and testing datasets, yielding coefficient of determination (R²) values of 0.9821 and 0.8169, and mean absolute error values of 0.0178 and 0.0108, respectively. Bayesian optimisation was employed using the multimodal model as a surrogate model to identify potential optimal combinations within the parameter space. These results validate the effectiveness of establishing a multimodal machine learning method based on multimodal data sources to optimise LPBF process parameters and improve relative density of fabricated new materials.

Limb motion mechanics analysis of Taijiquan trainees combined with APAS image analysis system

Analyzing the motion mechanics of Taijiquan trainees’ limbs can help trainees better understand the mechanics of Taijiquan movement and improve their training speed. However, all the current motion mechanics analysis models still have the disadvantage of poor motion mechanics analysis due to inaccurate motion image feature extraction. Therefore, this study utilizes the Ariel performance analysis system and temporal-difference to construct a motion mechanics analysis model to extract motion image features of Taijiquan and analyze its motion mechanics. analysis. The study first experimented with this analysis model. The outcomes indicated that the model achieved 100% accuracy in the extraction of image feature information and 100% accuracy in the analysis of motion mechanics, which was much higher than the comparative analysis model. The model was then used to analyze the motion mechanics of Taijiquan trainees during bending and squatting posture. The results revealed that the larger the knee fixation angle was, the shorter the completion time of the movement was. When the knee fixation angle was 150°, the time taken to complete the starting, fixing and finishing phases of the movement was 11.21 s, 61.21 s and 12.32 s respectively. Moreover, when the trainees performed the movement, the angle of the trainee’s right knee changed gently in the starting phase, while the angle of the trainee’s left knee changed more drastically. In summary, the motion mechanics analysis model proposed in the study is able to accurately analyze the limb motion mechanics of Taijiquan trainees as a means of avoiding various injuries during training.

Generative adversarial networks with texture recovery and physical constraints for remote sensing image dehazing

The scattering of tiny particles in the atmosphere causes a haze effect on remote sensing images captured by satellites and similar devices, significantly disrupting subsequent image recognition and classification. A generative adversarial network named TRPC-GAN with texture recovery and physical constraints is proposed to mitigate this impact. This network not only effectively removes haze but also better preserves the texture information of the original remote sensing image, thereby enhancing the visual quality of the dehazed image. A multi-scale module is proposed to extract feature information of remote sensing images, allowing it to capture image features from different receptive fields. Simultaneously, an attention module is designed further to guide the network’s focus towards important feature information. In addition, a multi-scale adversarial network is proposed to better restore both global and local information about the original image. Introducing a physical constraint loss function to improve the loss function of the original generative adversarial network allows for better preservation of the physical characteristics of remote sensing images. Simulation experiments on synthetic and natural hazy remote sensing image datasets are conducted. The results demonstrate that the dehazing performance of the TRPC-GAN method surpasses the other four methods.

Recent advances of machine learning in the geographical origin traceability of food and agro-products: A review.

The geographical origin traceability of food and agro-products has been attracted worldwide. Especially with the rise of machine learning (ML) technology, it provides cutting-edge solutions to erstwhile intractable issues to identify the origin of food and agro-products. By utilizing advanced algorithms, ML can extract feature information of food and agro-products that is closely related to origin and, more accurately, identify and trace their origins, which is of great significance to the entire food and agriculture industry. This paper provides a comprehensive overview of the state-of-the-art applications of ML in the geographical origin traceability of food and agro-products. First, commonly used ML methods are summarized. The paper then outlines the whole process of preparation for modeling, model training as well as model evaluation for building traceability models-based ML. Finally, recent applications of ML combined with different traceability techniques in the field of food and agro-products are revisited. Although ML has made many achievements in solving the geographical origin traceability problem of food and agro-products, it still has great development potential. For example, the application of ML is yet insufficient in the geographical origin traceability using DNA or computer vision techniques. The ability of ML to predict the geographical origin of food and agro-products can be further improved, for example, by increasing model interpretability, incorporating data fusion strategies, and others.

DCA-Enhanced Alzheimer's detection with shearlet and deep learning integration.

Alzheimer's dementia (AD) is a neurodegenerative disorder that affects the central nervous system, causing the cells to stop working or die. The quality of life for individuals with AD steadily declines over time. While current treatments can relieve symptoms, a definitive cure remains elusive. However, technological advancements in machine learning (ML) and deep learning (DL) have opened up new possibilities for early AD detection. Early diagnosis is crucial, as trial drugs show promising results in patients who are diagnosed early. This study used a magnetic resonance imaging (MRI) dataset from the Alzheimer's Disease Neuroimaging Initiative (ADNI). The dataset consisted of 200 patients who were followed up at different time points and categorized as having AD (50), progressive-mild cognitive impairment to AD (50), stable-mild cognitive impairment (50), or cognitively normal (50). However, the utilization of MRI datasets poses challenges such as high dimensionality, limited training samples, and variability within and between subjects. To overcome these challenges, I propose using convolutional neural networks (CNNs) to extract informative features from an MRI sample. I fine-tune four pretrained models (i.e., SqueezeNet-v1.1, MobileNet-v2, Xception, and Inception-v3) to generate discriminative descriptors of MRI sample characteristics. Additionally, I suggest using the 3D shearlet transform, considering the volumetric properties of MRI data. Before the transformation, I implemented preprocessing protocols such as skull stripping, normalization of image intensity, and spatial cropping. I then summarize the shearlet coefficients using texture-based techniques. Finally, I integrate both deep and shearlet-based features using discriminant correlation analysis (DCA) to yield a robust and computationally efficient classification model. I employ two classifiers, support vector machines (SVMs) and decision tree baggers (DTBs). My objective was to develop a model capable of accurately diagnosing early-stage AD that can facilitate effective intervention and management of the condition. Our feature representation demonstrated high accuracy when applied to AD datasets at three time points. Specifically, accuracies of 94.46%, 92.97%, and 95.44% were achieved 18 months, 12 months, and at the time of stable diagnosis, respectively.

RT-DETR-Tea: A Multi-Species Tea Bud Detection Model for Unstructured Environments

Accurate bud detection is a prerequisite for automatic tea picking and yield statistics; however, current research suffers from missed detection due to the variety of singleness and false detection under complex backgrounds. Traditional target detection models are mainly based on CNN, but CNN can only achieve the extraction of local feature information, which is a lack of advantages for the accurate identification of targets in complex environments, and Transformer can be a good solution to the problem. Therefore, based on a multi-variety tea bud dataset, this study proposes RT-DETR-Tea, an improved object detection model under the real-time detection Transformer (RT-DETR) framework. This model uses cascaded group attention to replace the multi-head self-attention (MHSA) mechanism in the attention-based intra-scale feature interaction (AIFI) module, effectively optimizing deep features and enriching the semantic information of features. The original cross-scale feature-fusion module (CCFM) mechanism is improved to establish the gather-and-distribute-Tea (GD-Tea) mechanism for multi-level feature fusion, which can effectively fuse low-level and high-level semantic information and large and small tea bud features in natural environments. The submodule of DilatedReparamBlock in UniRepLKNet was employed to improve RepC3 to achieve an efficient fusion of tea bud feature information and ensure the accuracy of the detection head. Ablation experiments show that the precision and mean average precision of the proposed RT-DETR-Tea model are 96.1% and 79.7%, respectively, which are increased by 5.2% and 2.4% compared to those of the original model, indicating the model’s effectiveness. The model also shows good detection performance on the newly constructed tea bud dataset. Compared with other detection algorithms, the improved RT-DETR-Tea model demonstrates superior tea bud detection performance, providing effective technical support for smart tea garden management and production.

Research progress on the depth of anesthesia monitoring based on the electroencephalogram

AbstractGeneral anesthesia typically involves three key components: amnesia, analgesia, and immobilization. Monitoring the depth of anesthesia (DOA) during surgery is crucial for personalizing anesthesia regimens and ensuring precise drug delivery. Since general anesthetics act primarily on the brain, this organ becomes the target for monitoring DOA. Electroencephalogram (EEG) can record the electrical activity generated by various brain tissues, enabling anesthesiologists to monitor the DOA from real‐time changes in a patient's brain activity during surgery. This monitoring helps to optimize anesthesia medication, prevent intraoperative awareness, and reduce the incidence of cardiovascular and other adverse events, contributing to anesthesia safety. Different anesthetic drugs exert different effects on the EEG characteristics, which have been extensively studied in commonly used anesthetic drugs. However, due to the limited understanding of the biological basis of consciousness and the mechanisms of anesthetic drugs acting on the brain, combined with the effects of various factors on existing EEG monitors, DOA cannot be accurately expressed via EEG. The lack of patient reactivity during general anesthesia does not necessarily indicate unconsciousness, highlighting the importance of distinguishing the mechanisms of consciousness and conscious connectivity when monitoring perioperative anesthesia depth. Although EEG is an important means of monitoring DOA, continuous optimization is necessary to extract characteristic information from EEG to monitor DOA, and EEG monitoring technology based on artificial intelligence analysis is an emerging research direction.

A novel spatio-temporal attention mechanism model for car-following in autonomous driving

Car following is critical for the overall safety, efficiency, and smooth operation of autonomous vehicles in traffic. However, existing car-following model primarily focuses on local feature extraction, overlooking the spatial–temporal relationships between vehicles and key information within the time series. This limitation negatively impacts prediction accuracy and generalization capabilities. To address these issues, a novel Spatio-Temporal Car-Following model based on Graph Convolution Network and Attention mechanism is proposed to enhance the safety, efficiency, and comfort of autonomous driving systems. Firstly, the spatial–temporal structure of the car-following model is constructed using the GCN network. This approach efficiently captures the topological relationships between vehicles. Secondly, we interleave spatial and temporal blocks to extract feature information from both spatial and temporal dimensions, which allows us to perceive spatial interactions and temporal motion patterns of the car. Additionally, a self-attention module assigns different attention weights to the various neighbors of a node in the car-following model, which can facilitate the capture of diverse aspects of node relationships and enhance expressive power. Finally, a Multi-Layer Perceptron (MLP) layer predicts the future behaviors of following vehicles. The proposed car-following model (CF) was trained and evaluated using five real-world datasets: HighD, SPMD, Waymo, NGSIM, and Lyft. The results indicate that our approach surpasses current models in terms of Mean Squared Error (MSE) for spacing, while also maintaining a zero collision rate across every dataset. These findings indicate that the Spatio-Temporal Attention Model (GSTAM-CF) proposed in this paper enhances safety, efficiency, and operational comfort compared to existing models.

A novel OL-mapping operator-based edge detection approach

Accurate edge detection is of great significance to image processing. However, the traditional methods, which use differential operations or prior experience to determine the value of the operator, cannot meet the high demand for accuracy in the field of image processing. Deep-learning based detection algorithms restrict its flexible usability due to the high demand of computational resources (storage space, processor) for its large number of parameters. To date, neither traditional nor emerging techniques cannot achieve SOTA performance on detection tasks. Therefore, to address the need for both precision and resources, this paper proposes an edge detection operator (HQ operator) based on the relationship between the original image and label mapping, which enhances the edge detection effect while drastically reducing the computational complexity. Unlike previous edge extraction operators, HQ operator utilizes a combination of labels and original images for edge detection, and does not use iterative training, but through subgraph generation classification (SGC) method and subgraph feature fusion operation (SFFO) to obtain the edge detection operator at once. Compared to the traditional edge detection operator, HQ operator can accurately extract feature information of the edge of target region. And compared to deep learning, the number of parameters of the HQ operator is greatly reduced and the calculation time of parameters is reduced. In addition, the HQ operator obtained on one dataset can be ported to other datasets with good edge extraction results. The edge images processed by the HQ operator can be spliced with the original image in the form of channels, and the image after splicing is fed into the image segmentation network to improve the performance of the segmentation network. to improve the performance of the segmentation network. The experimental results show that by using the HQ operator, the image edge detection effect is significantly improved (PA:96.49%), and the performance of the segmentation network is also improved to a certain extent when this operator is combined with the segmentation network(Dice:89.97%, IoU:81.97%).

A Survey of Fine-Grained Visual Categorization Based on Deep Learning

Deep learning has achieved excellent results in various tasks in the field of computer vision, especially in fine-grained visual categorization. It aims to distinguish the subordinate categories of the label-level categories. Due to high intra-class variances and high inter-class similarity, the fine-grained visual categorization is extremely challenging. This paper first briefly introduces and analyzes the related public datasets. After that, some of the latest methods are reviewed. Based on the feature types, the feature processing methods, and the overall structure used in the model, we divide them into three types of methods: methods based on general convolutional neural network (CNN) and strong supervision of parts, methods based on single feature processing, and methods based on multiple feature processing. Most methods of the first type have a relatively simple structure, which is the result of the initial research. The methods of the other two types include models that have special structures and training processes, which are helpful to obtain discriminative features. We conduct a specific analysis on several methods with high accuracy on public datasets. In addition, we support that the focus of the future research is to solve the demand of existing methods for the large amount of the data and the computing power. In terms of technology, the extraction of the subtle feature information with the burgeoning vision transformer (ViT) network is also an important research direction.