Spatial Pyramid Pooling Research Articles

Multi-class segmentation of navel orange surface defects based on improved DeepLabv3+

To address the problems of current mainstream semantic segmentation network such as rough edge segmentation of navel oranges defects, poor accuracy of small target defect segmentation and insufficient deep-level semantic extraction of defects, feature information will be lost, a multi-class segmentation model based on improved DeepLabv3+ is proposed to detect the surface defects of navel oranges. The Coordinate Attention Mechanism is embedded into the DeepLabv3+ network for better semantic segmentation performance, while the dilated convolution of Atrous Spatial Pyramid Pooling structure is replaced with deformable empty convolution to improve the fitting ability of the network to target shape changes and irregular defects. In addition, a BiFPN-based feature fusion branch is introduced at the DeepLabv3+ encoder side to realize multi-scale feature fusion and enrich feature space and semantic information. The experimental results show that the average intersection ratio and average pixel intersection ratio accuracies of the improved DeepLabv3+ model on the navel orange surface defect dataset are 77.32% and 86.38%, which are 3.81% and 5.29% higher than the original DeepLabv3+ network, respectively, improving the extraction capability of navel orange defect features and having better segmentation performance.

Urban Vegetation Classification for Unmanned Aerial Vehicle Remote Sensing Combining Feature Engineering and Improved DeepLabV3+

Addressing the problems of misclassification and omissions in urban vegetation fine classification from current remote sensing classification methods, this research proposes an intelligent urban vegetation classification method that combines feature engineering and improved DeepLabV3+ based on unmanned aerial vehicle visible spectrum images. The method constructs feature engineering under the ReliefF algorithm to increase the number of features in the samples, enabling the deep learning model to learn more detailed information about the vegetation. Moreover, the method improves the classical DeepLabV3+ network structure based on (1) replacing the backbone network using MoblieNetV2; (2) adjusting the atrous spatial pyramid pooling null rate; and (3) adding the attention mechanism and the convolutional block attention module. Experiments were conducted with self-constructed sample datasets, where the method was compared and analyzed with a fully convolutional network (FCN) and U-Net and ShuffleNetV2 networks; the migration of the method was tested as well. The results show that the method in this paper is better than FCN, U-Net, and ShuffleNetV2, and reaches 92.27%, 91.48%, and 85.63% on the accuracy evaluation indices of overall accuracy, MarcoF1, and mean intersection over union, respectively. Furthermore, the segmentation results are accurate and complete, which effectively alleviates misclassifications and omissions of urban vegetation; moreover, it has a certain migration ability that can quickly and accurately classify the vegetation.

Ancient mural segmentation based on multiscale feature fusion and dual attention enhancement

To address the fuzzy segmentation boundaries, missing details, small target losses and low efficiency of traditional segmentation methods in ancient mural image segmentation scenarios, this paper proposes a mural segmentation model based on multiscale feature fusion and a dual attention-augmented segmentation model (MFAM). The model uses the MobileViT network, which integrates a coordinate attention mechanism, as the feature extraction backbone network. It attains global and local expression capabilities through self-attention, class convolution, and coordinate attention and focuses on location information to expand the receptive field and achieve improved feature extraction efficiency. An A_R_ASPP feature enhancement module is proposed for the attention-optimized residual atrous spatial pyramid pooling module. The module uses residual connections to solve the small target loss problem in murals caused by the excessive sampling rate of atrous convolution and uses a feature attention mechanism to adaptively adjust the feature map weight according to the channel importance levels. A dual attention-enhanced feature fusion module is proposed for multiscale decoder feature fusion to improve the mural segmentation effect. This module uses a cross-level aggregation strategy and an attention mechanism to weight the importance of different feature levels to obtain multilevel semantic feature representations. The model improves the mean intersection over union (MIoU) by 3.06% and the MPA by 1.81% on a mural dataset compared with other models. The model is proven to be effective at improving the segmentation details, efficiency and small target segmentation results produced for mural images, and a new method is proposed for segmenting ancient mural images.

Detection of coconut leaf diseases using enhanced deep learning techniques

Coconut farming is a significant agricultural activity in South India, but the coconut trees face challenges due to adverse weather conditions and environmental factors. These challenges include various leaf diseases and pest infestations. Identifying and locating these issues can be difficult because of the large foliage and shading provided by the coconut trees. Recent research has shown that Computer Vision algorithms are becoming increasingly important for solving problems related to object identification and detection. So, in this work, the YOLOv4 algorithm was employed to detect and pinpoint diseases and infections in coconut leaves from images. The YOLOv4 model incorporates advanced features such as cross-stage partial connections, spatial pyramid pooling, contextual feature selection, and path-based aggregation. These features enhance the model’s ability to efficiently identify issues such as yellowing and drying of leaves, pest infections, and leaf flaccidity in coconut leaf images taken in various environmental conditions. Furthermore, the model’s predictive accuracy was enhanced through multi-scale feature detection, PANet feature learning, and adaptive bounding boxes. These improvements resulted in an impressive 88% F1-Score and an 85% Mean Average Precision. The model demonstrates its effectiveness and robustness even when dealing with medium-resolution images, offering improved accuracy and speed in disease and pest detection on coconut leaves.

Deep CNN-based microaneurysm segmentation system in retinal images using multi-level features

Microaneurysms, tiny, circular red dots that occur in retinal fundus images, are one of the earliest symptoms of diabetic retinopathy. Because microaneurysms are small and delicate, detecting them can be difficult. Their small size and cunning character make automatic detection of them difficult. In this study, a novel encoder-decoder network is proposed to segment the MAs automatically and accurately. The encoder part mainly consists of three parts: a low-level feature extraction module composed of a dense connectivity block (Dense Block), a High-resolution Block (HR Block), and an Atrous Spatial Pyramid Pooling (ASPP) module, of which the latter two modules are used to extract high-level information. Therefore, the network is named a Multi-Level Features based Deep Convolutional Neural Network (MF-DCNN). The proposed decoder takes advantage of the multi-scale features from the encoder to predict MA regions. Compared with the existing methods on three datasets, it is proved that the proposed method is better than the current excellent methods in the segmentation results of the normal and abnormal fundus. In the case of fewer network parameters, MF-DCNN achieves better prediction performance on intersection over union (IoU), dice similarity coefficient (DSC), and other evaluation metrics. MF-DCNN is lightweight and able to use multi-scale features to predict MA regions. It can be used to automatically segment the MA and assist in computer-aided diagnosis.

Lightweight high-precision SAR ship detection method based on YOLOv7-LDS.

The current challenges in Synthetic Aperture Radar (SAR) ship detection tasks revolve around handling significant variations in target sizes and managing high computational expenses, which hinder practical deployment on satellite or mobile airborne platforms. In response to these challenges, this research presents YOLOv7-LDS, a lightweight yet highly accurate SAR ship detection model built upon the YOLOv7 framework. In the core of YOLOv7-LDS's architecture, we introduce a streamlined feature extraction network that strikes a delicate balance between detection precision and computational efficiency. This network is founded on Shufflenetv2 and incorporates Squeeze-and-Excitation (SE) attention mechanisms as its key elements. Additionally, in the Neck section, we introduce the Weighted Efficient Aggregation Network (DCW-ELAN), a fundamental feature extraction module that leverages Coordinate Attention (CA) and Depthwise Convolution (DWConv). This module efficiently aggregates features while preserving the ability to identify small-scale variations, ensuring top-quality feature extraction. Furthermore, we introduce a lightweight Spatial Pyramid Dilated Convolution Cross-Stage Partial Channel (LSPHDCCSPC) module. LSPHDCCSPC is a condensed version of the Spatial Pyramid Pooling Cross-Stage Partial Channel (SPPCSPC) module, incorporating Dilated Convolution (DConv) as a central component for extracting multi-scale information. The experimental results show that YOLOv7-LDS achieves a remarkable Mean Average Precision (mAP) of 99.1% and 95.8% on the SAR Ship Detection Dataset (SSDD) and the NWPU VHR-10 dataset with a parameter count (Params) of 3.4 million, a Giga Floating Point Operations Per Second (GFLOPs) of 6.1 and an Inference Time (IT) of 4.8 milliseconds. YOLOv7-LDS effectively strikes a fine balance between computational cost and detection performance, surpassing many of the current state-of-the-art object detection models. As a result, it offers a more resilient solution for maritime ship monitoring.

A Novel Crop Pest Detection Model Based on YOLOv5

The damage caused by pests to crops results in reduced crop yield and compromised quality. Accurate and timely pest detection plays a crucial role in helping farmers to defend against and control pests. In this paper, a novel crop pest detection model named YOLOv5s-pest is proposed. Firstly, we design a hybrid spatial pyramid pooling fast (HSPPF) module, which enhances the model’s capability to capture multi-scale receptive field information. Secondly, we design a new convolutional block attention module (NCBAM) that highlights key features, suppresses redundant features, and improves detection precision. Thirdly, the recursive gated convolution (g3Conv) is introduced into the neck, which extends the potential of self-attention mechanism to explore feature representation to arbitrary-order space, enhances model capacity and detection capability. Finally, we replace the non-maximum suppression (NMS) in the post-processing part with Soft-NMS, which improves the missed problem of detection in crowded and dense scenes. The experimental results show that the mAP@0.5 (mean average precision at intersection over union (IoU) threshold of 0.5) of YOLOv5s-pest achieves 92.5% and the mAP@0.5:0.95 (mean average precision from IoU 0.5 to 0.95) achieves 72.6% on the IP16. Furthermore, we also validate our proposed method on other datasets, and the outcomes indicate that YOLOv5s-pest is also effective in other detection tasks.

ASD-Net: a novel U-Net based asymmetric spatial-channel convolution network for precise kidney and kidney tumor image segmentation

Early intervention in tumors can greatly improve human survival rates. With the development of deep learning technology, automatic image segmentation has taken a prominent role in the field of medical image analysis. Manually segmenting kidneys on CT images is a tedious task, and due to the diversity of these images and varying technical skills of professionals, segmentation results can be inconsistent. To address this problem, a novel ASD-Net network is proposed in this paper for kidney and kidney tumor segmentation tasks. First, the proposed network employs newly designed Adaptive Spatial-channel Convolution Optimization (ASCO) blocks to capture anisotropic information in the images. Then, other newly designed blocks, i.e., Dense Dilated Enhancement Convolution (DDEC) blocks, are utilized to enhance feature propagation and reuse it across the network, thereby improving its segmentation accuracy. To allow the network to segment complex and small kidney tumors more effectively, the Atrous Spatial Pyramid Pooling (ASPP) module is incorporated in its middle layer. With its generalized pyramid feature, this module enables the network to better capture and understand context information at various scales within the images. In addition to this, the concurrent spatial and channel squeeze & excitation (scSE) attention mechanism is adopted to better comprehend and manage context information in the images. Additional encoding layers are also added to the base (U-Net) and connected to the original encoding layer through skip connections. The resultant enhanced U-Net structure allows for better extraction and merging of high-level and low-level features, further boosting the network’s ability to restore segmentation details. In addition, the combined Binary Cross Entropy (BCE)-Dice loss is utilized as the network's loss function. Experiments, conducted on the KiTS19 dataset, demonstrate that the proposed ASD-Net network outperforms the existing segmentation networks according to all evaluation metrics used, except for recall in the case of kidney tumor segmentation, where it takes the second place after Attention-UNet.Graphical

Real-time semantic segmentation network with an enhanced backbone based on Atrous spatial pyramid pooling module

Real-time semantic segmentation methods demand high levels of both speed and accuracy. However, recent methods in semantic segmentation generally use the models pre-trained on ImageNet as their backbone to increase the field-of-view quickly. Due to the low resolution and plenty of image categories, the existing methods often suffer from insufficient receptive field and channel redundancy. To tackle these problems, we propose a novel block structure named Atrous block (A block) in this paper. By repeating this block structure, we have developed our dedicated backbone tailored for the semantic segmentation task, eliminating the need for additional modules. According to the different output channels of the backbone, we obtain our own architecture called Atrous Network (ANet), Furthermore, we propose a lightweight decoder to recover the spatial details lost during encoding. To demonstrate the effectiveness of our methods, we have validated our methods on CityScapes and CamVid datasets, using mixed precision during the training process. Compared with the state-of-the-art methods, our models achieve competitive results. Among them, our small size model achieves 75.2 % mean intersection over union (mIOU) on CityScapes dataset and 78.3 % mIOU on CamVid dataset, both without ImageNet pretraining.

Attention-Based Monocular Depth Estimation Considering Global and Local Information in Remote Sensing Images

Monocular depth estimation using a single remote sensing image has emerged as a focal point in both remote sensing and computer vision research, proving crucial in tasks such as 3D reconstruction and target instance segmentation. Monocular depth estimation does not require multiple views as references, leading to significant improvements in both time and efficiency. Due to the complexity, occlusion, and uneven depth distribution of remote sensing images, there are currently few monocular depth estimation methods for remote sensing images. This paper proposes an approach to remote sensing monocular depth estimation that integrates an attention mechanism while considering global and local feature information. Leveraging a single remote sensing image as input, the method outputs end-to-end depth estimation for the corresponding area. In the encoder, the proposed method employs a dense neural network (DenseNet) feature extraction module with efficient channel attention (ECA), enhancing the capture of local information and details in remote sensing images. In the decoder stage, this paper proposes a dense atrous spatial pyramid pooling (DenseASPP) module with channel and spatial attention modules, effectively mitigating information loss and strengthening the relationship between the target’s position and the background in the image. Additionally, weighted global guidance plane modules are introduced to fuse comprehensive features from different scales and receptive fields, finally predicting monocular depth for remote sensing images. Extensive experiments on the publicly available WHU-OMVS dataset demonstrate that our method yields better depth results in both qualitative and quantitative metrics.

A Building Extraction Method for High-Resolution Remote Sensing Images with Multiple Attentions and Parallel Encoders Combining Enhanced Spectral Information

Accurately extracting pixel-level buildings from high-resolution remote sensing images is significant for various geographical information applications. Influenced by different natural, cultural, and social development levels, buildings may vary in shape and distribution, making it difficult for the network to maintain a stable segmentation effect of buildings in different areas of the image. In addition, the complex spectra of features in remote sensing images can affect the extracted details of multi-scale buildings in different ways. To this end, this study selects parts of Xi’an City, Shaanxi Province, China, as the study area. A parallel encoded building extraction network (MARS-Net) incorporating multiple attention mechanisms is proposed. MARS-Net builds its parallel encoder through DCNN and transformer to take advantage of their extraction of local and global features. According to the different depth positions of the network, coordinate attention (CA) and convolutional block attention module (CBAM) are introduced to bridge the encoder and decoder to retain richer spatial and semantic information during the encoding process, and adding the dense atrous spatial pyramid pooling (DenseASPP) captures multi-scale contextual information during the upsampling of the layers of the decoder. In addition, a spectral information enhancement module (SIEM) is designed in this study. SIEM further enhances building segmentation by blending and enhancing multi-band building information with relationships between bands. The experimental results show that MARS-Net performs better extraction results and obtains more effective enhancement after adding SIEM. The IoU on the self-built Xi’an and WHU building datasets are 87.53% and 89.62%, respectively, while the respective F1 scores are 93.34% and 94.52%.

GIFNet: an effective global infection feature network for automatic COVID-19 lung lesions segmentation.

The ongoing COronaVIrus Disease 2019 (COVID-19) pandemic carried by the SARS-CoV-2 virus spread worldwide in early 2019, bringing about an existential health catastrophe. Automatic segmentation of infected lungs from COVID-19 X-ray and computer tomography (CT) images helps to generate a quantitative approach for treatment and diagnosis. The multi-class information about the infected lung is often obtained from the patient's CT dataset. However, the main challenge is the extensive range of infected features and lack of contrast between infected and normal areas. To resolve these issues, a novel Global Infection Feature Network (GIFNet)-based Unet with ResNet50 model is proposed for segmenting the locations of COVID-19 lung infections. The Unet layers have been used to extract the features from input images and select the region of interest (ROI) by using the ResNet50 technique for training it faster. Moreover, integrating the pooling layer into the atrous spatial pyramid pooling (ASPP) mechanism in the bottleneck helps for better feature selection and handles scale variation during training. Furthermore, the partial differential equation (PDE) approach is used to enhance the image quality and intensity value for particular ROI boundary edges in the COVID-19 images. The proposed scheme has been validated on two datasets, namely the SARS-CoV-2 CT scan and COVIDx-19, for detecting infected lung segmentation (ILS). The experimental findings have been subjected to a comprehensive analysis using various evaluation metrics, including accuracy (ACC), area under curve (AUC), recall (REC), specificity (SPE), dice similarity coefficient (DSC), mean absolute error (MAE), precision (PRE), and mean squared error (MSE) to ensure rigorous validation. The results demonstrate the superior performance of the proposed system compared to the state-of-the-art (SOTA) segmentation models on both X-ray and CT datasets.

Optical gas imaging for leak detection based on improved deeplabv3+ model

Optical gas imaging (OGI) is a technique to dynamically observe gas leaks with a wide detection range. As gasses are mobile and diffusive, traditional motion detection methods are difficult to adapt to the change of gas motion state with the environment, resulting in poor segmentation effect and low accuracy. Therefore, this paper proposes a deep learning-based semantic segmentation method of gas for optical gas imaging detection based on DeeplabV3+ model. Firstly, MobileNetV2 is used to replace the original backbone network to reduced model parameters and complexity and facilitates the edge deployment of the model. Second, the dense atrous spatial pyramid pooling (DenseASPP) module is introduced to improve the atrous spatial pyramid pooling (ASPP) module of the model to increase the model sensing field and improve the model gas segmentation accuracy. Finally, Dice loss is used to solve the sample imbalance problem. The experimental results show that IOU and F1 score of the model are improved by 4.10% and 2.47%, and it provides a new intelligent detection method for optical gas imaging.

FPIRST: Fatigue Driving Recognition Method Based on Feature Parameter Images and a Residual Swin Transformer.

Fatigue driving is a serious threat to road safety, which is why accurately identifying fatigue driving behavior and warning drivers in time are of great significance in improving traffic safety. However, accurately recognizing fatigue driving is still challenging due to large intra-class variations in facial expression, continuity of behaviors, and illumination conditions. A fatigue driving recognition method based on feature parameter images and a residual Swin Transformer is proposed in this paper. First, the face region is detected through spatial pyramid pooling and a multi-scale feature output module. Then, a multi-scale facial landmark detector is used to locate 23 key points on the face. The aspect ratios of the eyes and mouth are calculated based on the coordinates of these key points, and a feature parameter matrix for fatigue driving recognition is obtained. Finally, the feature parameter matrix is converted into an image, and the residual Swin Transformer network is presented to recognize fatigue driving. Experimental results on the HNUFD dataset show that the proposed method achieves an accuracy of 96.512%, thus outperforming state-of-the-art methods.

Research on the application of lightweight deep learning model in automatic identification of electricity-related public safety hazards in distribution lines

Abstract In view of the current problems of low speed, high network complexity, and difficulty in accurately detecting small target defects in insulator defect detection methods, this study proposes a lightweight insulator defect detection model. First, by enhancing both the convolutional block (CBL) and the efficient long-range aggregation network (ELAN-S) within the feature extraction network, the extraction capability for defect features is significantly augmented. Secondly, the CA-Tiny Spatial Pyramid Pooling (SPP) module was crafted by seamlessly integrating the coordinate attention mechanism with Tiny SPP, enabling the model to prioritize insulator defect characteristics and thereby enhancing defect detection accuracy. Finally, utilizing the positioning loss function WIoUv3 loss for loss calculation, a smaller gradient gain is assigned to low-quality anchor boxes, minimizing harmful gradients and enhancing the model’s positioning performance. Experimental outcomes demonstrate that the enhanced YOLOv7-tiny model excels in rapid and precise defect detection. In comparison to the original YOLOv7-tiny model, the proposed version is well-suited for edge device deployment, enabling real-time insulator defect detection.

MSFF: An automated multi-scale feature fusion deep learning model for spine fracture segmentation using MRI

Spinal fractures pose a significant healthcare challenge, impacting patient care and clinical decisions. Accurately identifying these fractures from medical images is crucial due to their location, shape, type, and severity. Meeting these challenges often necessitates employing advanced machine learning and deep learning techniques. This study introduces a novel deep learning model called MSFF (Multi-Scale Feature Fusion) specifically designed for automated spine fracture segmentation using MRI images. The MSFF architecture comprises five key modules: the Feature Fusion Module (FFM), Squeeze and Excitation (SEM), Atrous Spatial Pyramid Pooling (ASPP), Residual Convolution Block Attention Module (RCBAM), Residual Border Refinement Attention Block (RBRAB), and Local Position Residual Attention Block (LPRAB). These modules collectively facilitate multi-scale feature fusion, spatial feature extraction, channel-wise feature enhancement, segmentation border refinement, and focused attention on the region of interest. Subsequently, a decoder network is employed to predict fractured spine areas. Experimental findings demonstrate that the proposed MSFF approach attains superior accuracy in addressing the challenges and outperforms existing segmentation methods.

Research on Coal and Gangue Recognition Based on the Improved YOLOv7-Tiny Target Detection Algorithm.

The recognition technology of coal and gangue is one of the key technologies of intelligent mine construction. Aiming at the problems of the low accuracy of coal and gangue recognition models and the difficult recognition of small-target coal and gangue caused by low-illumination and high-dust environments in the coal mine working face, a coal and gangue recognition model based on the improved YOLOv7-tiny target detection algorithm is proposed. This paper proposes three model improvement methods. The coordinate attention mechanism is introduced to improve the feature expression ability of the model. The contextual transformer module is added after the spatial pyramid pooling structure to improve the feature extraction ability of the model. Based on the idea of the weighted bidirectional feature pyramid, the four branch modules in the high-efficiency layer aggregation network are weighted and cascaded to improve the recognition ability of the model for useful features. The experimental results show that the average precision mean of the improved YOLOv7-tiny model is 97.54%, and the FPS is 24.73 f·s-1. Compared with the Faster-RCNN, YOLOv3, YOLOv4, YOLOv4-VGG, YOLOv5s, YOLOv7, and YOLOv7-tiny models, the improved YOLOv7-tiny model has the highest recognition rate and the fastest recognition speed. Finally, the improved YOLOv7-tiny model is verified by field tests in coal mines, which provides an effective technical means for the accurate identification of coal and gangue.

Dual‐Stream Fusion and Multi‐scale Analysis: Introducing the Synergistic Dual‐Stream Network (SDS‐Net) for Image Manipulation Segmentation

This article introduces the synergistic dual‐stream network (SDS‐Net), a novel neural network architecture that significantly enhances the detection of image manipulations. SDS‐Net employs a unique dual‐stream fusion strategy that processes both RGB image and the corresponding noise map. It innovatively combines features computation from the different blocks of dual backbones, and leverages a multi‐scale spatial pyramid pooling (MSPP) module to expand the receptive fields of shallow features. This approach not only enriches the feature representation, but also ensures the precise localization of manipulated regions. Extensive experiments conducted on various public datasets demonstrate the superiority of SDS‐Net over several state‐of‐the‐art methods.

Enhanced YOLO Network for Improving the Efficiency of Traffic Sign Detection

One important task for autonomous driving is the precise detection and recognition of road traffic signs. This research focuses on a comprehensive set of 72 distinct traffic signs that are prevalent on urban roads in China, with the goal of developing an enhanced You Only Look Once (YOLO) network model tailored for this specific task. The modifications include the omission of the terminal convolution module and Conv3 (C3) module within the backbone network. Additionally, the 32-fold downsampling is replaced with a 16-fold downsampling, and a feature fusion module with dimensions of 152 × 152 is introduced in the feature layer. To capture a more encompassing context, a novel hybrid space pyramid pooling module, referred to as Hybrid Spatial Pyramid Pooling Fast (H-SPPF), is introduced. Furthermore, a channel attention mechanism is integrated into the framework, combined with three other improved methodologies. Upon evaluation, the enhanced algorithm demonstrates impressive results, achieving a precision rate of 91.72%, a recall rate of 91.77%, and a mean average precision (mAP) of 93.88% at an intersection over union (IoU) threshold of 0.5. Additionally, the method also achieves an mAP of 75.81% for a variety of IoU criteria between 0.5 and 0.95. These achievements are validated on an augmented dataset established for this study.

Research on the Construction of an Efficient and Lightweight Online Detection Method for Tiny Surface Defects through Model Compression and Knowledge Distillation

In response to the current issues of poor real-time performance, high computational costs, and excessive memory usage of object detection algorithms based on deep convolutional neural networks in embedded devices, a method for improving deep convolutional neural networks based on model compression and knowledge distillation is proposed. Firstly, data augmentation is employed in the preprocessing stage to increase the diversity of training samples, thereby improving the model’s robustness and generalization capability. The K-means++ clustering algorithm generates candidate bounding boxes, adapting to defects of different sizes and selecting finer features earlier. Secondly, the cross stage partial (CSP) Darknet53 network and spatial pyramid pooling (SPP) module extract features from the input raw images, enhancing the accuracy of defect location detection and recognition in YOLO. Finally, the concept of model compression is integrated, utilizing scaling factors in the batch normalization (BN) layer, and introducing sparse factors to perform sparse training on the network. Channel pruning and layer pruning are applied to the sparse model, and post-processing methods using knowledge distillation are used to effectively reduce the model size and forward inference time while maintaining model accuracy. The improved model size decreases from 244 M to 4.19 M, the detection speed increases from 32.8 f/s to 68 f/s, and mAP reaches 97.41. Experimental results demonstrate that this method is conducive to deploying network models on embedded devices with limited GPU computing and storage resources. It can be applied in distributed service architectures for edge computing, providing new technological references for deploying deep learning models in the industrial sector.