Enhanced Lightweight YOLOX for Small Object Wildfire Detection in UAV Imagery.

Target detection technology based on unmanned aerial vehicle (UAV)-derived aerial imagery has been widely applied in the field of forest fire patrol and rescue. However, due to the specificity of UAV platforms, there are still significant issues to be resolved such as severe omission, low detection accuracy, and poor early warning effectiveness. In light of these issues, this paper proposes an improved YOLOX network for the rapid detection of forest fires in images captured by UAVs. Firstly, to enhance the network's feature-extraction capability in complex fire environments, a multi-level-feature-extraction structure, CSP-ML, is designed to improve the algorithm's detection accuracy for small-target fire areas. Additionally, a CBAM attention mechanism is embedded in the neck network to reduce interference caused by background noise and irrelevant information. Secondly, an adaptive-feature-extraction module is introduced in the YOLOX network's feature fusion part to prevent the loss of important feature information during the fusion process, thus enhancing the network's feature-learning capability. Lastly, the CIoU loss function is used to replace the original loss function, to address issues such as excessive optimization of negative samples and poor gradient-descent direction, thereby strengthening the network's effective recognition of positive samples. Experimental results show that the improved YOLOX network has better detection performance, with mAP@50 and mAP@50_95 increasing by 6.4% and 2.17%, respectively, compared to the traditional YOLOX network. In multi-target flame and small-target flame scenarios, the improved YOLO model achieved a mAP of 96.3%, outperforming deep learning algorithms such as FasterRCNN, SSD, and YOLOv5 by 33.5%, 7.7%, and 7%, respectively. It has a lower omission rate and higher detection accuracy, and it is capable of handling small-target detection tasks in complex fire environments. This can provide support for UAV patrol and rescue applications from a high-altitude perspective.

An MRS-YOLO Model for High-Precision Waste Detection and Classification.

With the advancement in living standards, there has been a significant surge in the quantity and diversity of household waste. To safeguard the environment and optimize resource utilization, there is an urgent demand for effective and cost-efficient intelligent waste classification methodologies. This study presents MRS-YOLO (Multi-Resolution Strategy-YOLO), a waste detection and classification model. The paper introduces the SlideLoss_IOU technique for detecting small objects, integrates RepViT of the Transformer mechanism, and devises a novel feature extraction strategy by amalgamating multi-dimensional and dynamic convolution mechanisms. These enhancements not only elevate the detection accuracy and speed but also bolster the robustness of the current YOLO model. Validation conducted on a dataset comprising 12,072 samples across 10 categories, including recyclable metal and paper, reveals a 3.6% enhancement in mAP50% accuracy compared to YOLOv8, coupled with a 15.09% reduction in volume. Furthermore, the model demonstrates improved accuracy in detecting small targets and exhibits comprehensive detection capabilities across diverse scenarios. For transparency and to facilitate further research, the source code and related datasets used in this study have been made publicly available at GitHub.

SEB-YOLO: An Improved YOLOv5 Model for Remote Sensing Small Target Detection.

Due to the limited semantic information extraction with small objects and difficulty in distinguishing similar targets, it brings great challenges to target detection in remote sensing scenarios, which results in poor detection performance. This paper proposes an improved YOLOv5 remote sensing image target detection algorithm, SEB-YOLO (SPD-Conv + ECSPP + Bi-FPN + YOLOv5). Firstly, the space-to-depth (SPD) layer followed by a non-strided convolution (Conv) layer module (SPD-Conv) was used to reconstruct the backbone network, which retained the global features and reduced the feature loss. Meanwhile, the pooling module with the attention mechanism of the final layer of the backbone network was designed to help the network better identify and locate the target. Furthermore, a bidirectional feature pyramid network (Bi-FPN) with bilinear interpolation upsampling was added to improve bidirectional cross-scale connection and weighted feature fusion. Finally, the decoupled head is introduced to enhance the model convergence and solve the contradiction between the classification task and the regression task. Experimental results on NWPU VHR-10 and RSOD datasets show that the mAP of the proposed algorithm reaches 93.5% and 93.9%respectively, which is 4.0% and 5.3% higher than that of the original YOLOv5l algorithm. The proposed algorithm achieves better detection results for complex remote sensing images.

MWIRGas-YOLO: Gas Leakage Detection Based on Mid-Wave Infrared Imaging.

The integration of visual algorithms with infrared imaging technology has become an effective tool for industrial gas leak detection. However, existing research has mostly focused on simple scenarios where a gas plume is clearly visible, with limited studies on detecting gas in complex scenes where target contours are blurred and contrast is low. This paper uses a cooled mid-wave infrared (MWIR) system to provide high sensitivity and fast response imaging and proposes the MWIRGas-YOLO network for detecting gas leaks in mid-wave infrared imaging. This network effectively detects low-contrast gas leakage and segments the gas plume within the scene. In MWIRGas-YOLO, it utilizes the global attention mechanism (GAM) to fully focus on gas plume targets during feature fusion, adds a small target detection layer to enhance information on small-sized targets, and employs transfer learning of similar features from visible light smoke to provide the model with prior knowledge of infrared gas features. Using a cooled mid-wave infrared imager to collect gas leak images, the experimental results show that the proposed algorithm significantly improves the performance over the original model. The segment mean average precision reached 96.1% (mAP50) and 47.6% (mAP50:95), respectively, outperforming the other mainstream algorithms. This can provide an effective reference for research on infrared imaging for gas leak detection.

Underwater Rescue Target Detection Based on Acoustic Images.

In order to effectively respond to floods and water emergencies that result in the drowning of missing persons, timely and effective search and rescue is a very critical step in underwater rescue. Due to the complex underwater environment and low visibility, unmanned underwater vehicles (UUVs) with sonar are more efficient than traditional manual search and rescue methods to conduct active searches using deep learning algorithms. In this paper, we constructed a sound-based rescue target dataset that encompasses both the source and target domains using deep transfer learning techniques. For the underwater acoustic rescue target detection of small targets, which lack image feature accuracy, this paper proposes a two-branch convolution module and improves the YOLOv5s algorithm model to design an acoustic rescue small target detection algorithm model. For an underwater rescue target dataset based on acoustic images with a small sample acoustic dataset, a direct fine-tuning using optical image pre-training lacks cross-domain adaptability due to the different statistical properties of optical and acoustic images. This paper therefore proposes a heterogeneous information hierarchical migration learning method. For the false detection of acoustic rescue targets in a complex underwater background, the network layer is frozen during the hierarchical migration of heterogeneous information to improve the detection accuracy. In addition, in order to be more applicable to the embedded devices carried by underwater UAVs, an underwater acoustic rescue target detection algorithm based on ShuffleNetv2 is proposed to improve the two-branch convolutional module and the backbone network of YOLOv5s algorithm, and to create a lightweight model based on hierarchical migration of heterogeneous information. Through extensive comparative experiments conducted on various acoustic images, we have thoroughly validated the feasibility and effectiveness of our method. Our approach has demonstrated state-of-the-art performance in underwater search and rescue target detection tasks.

Lightweight SM-YOLOv5 Tomato Fruit Detection Algorithm for Plant Factory.

Due to their rapid development and wide application in modern agriculture, robots, mobile terminals, and intelligent devices have become vital technologies and fundamental research topics for the development of intelligent and precision agriculture. Accurate and efficient target detection technology is required for mobile inspection terminals, picking robots, and intelligent sorting equipment in tomato production and management in plant factories. However, due to the limitations of computer power, storage capacity, and the complexity of the plant factory (PF) environment, the precision of small-target detection for tomatoes in real-world applications is inadequate. Therefore, we propose an improved Small MobileNet YOLOv5 (SM-YOLOv5) detection algorithm and model based on YOLOv5 for target detection by tomato-picking robots in plant factories. Firstly, MobileNetV3-Large was used as the backbone network to make the model structure lightweight and improve its running performance. Secondly, a small-target detection layer was added to improve the accuracy of small-target detection for tomatoes. The constructed PF tomato dataset was used for training. Compared with the YOLOv5 baseline model, the mAP of the improved SM-YOLOv5 model was increased by 1.4%, reaching 98.8%. The model size was only 6.33 MB, which was 42.48% that of YOLOv5, and it required only 7.6 GFLOPs, which was half that required by YOLOv5. The experiment showed that the improved SM-YOLOv5 model had a precision of 97.8% and a recall rate of 96.7%. The model is lightweight and has excellent detection performance, and so it can meet the real-time detection requirements of tomato-picking robots in plant factories.

DYNet: A Printed Book Detection Model Using Dual Kernel Neural Networks.

Target detection has always been a hotspot in image processing/computer vision research, and small-target detection is a frequently encountered problem in the field of target detection. With the continuous innovation of target detection technology, people always hope that the detection of small targets can reach the real-time accuracy of large-target detection. In this paper, a small-target detection model based on dual-core convolutional neural networks (CNN) is proposed, which is mainly used for the intelligent detection of books in the production line of printed books. The model is mainly composed of two modules, including a region prediction module and suspicious target search module. The region prediction module uses a CNN to predict suspicious region blocks in a large context. The suspicious target search module uses a different CNN from the above to find tiny targets in the predicted region blocks. Comparative testing of four small book target samples using this model shows that this model has better book small-target detection accuracy compared to other models.

APM-YOLOv7 for Small-Target Water-Floating Garbage Detection Based on Multi-Scale Feature Adaptive Weighted Fusion.

As affected by limited information and the complex background, the accuracy of small-target water-floating garbage detection is low. To increase the detection accuracy, in this research, a small-target detection method based on APM-YOLOv7 (the improved YOLOv7 with ACanny PConv-ELAN and MGA attention) is proposed. Firstly, the adaptive algorithm ACanny (adaptive Canny) for river channel outline extraction is proposed to extract the river channel information from the complex background, mitigating interference of the complex background and more accurately extracting the features of small-target water-floating garbage. Secondly, the lightweight partial convolution (PConv) is introduced, and the partial convolution-efficient layer aggregation network module (PConv-ELAN) is designed in the YOLOv7 network to improve the feature extraction capability of the model from morphologically variable water-floating garbage. Finally, after analyzing the limitations of the YOLOv7 network in small-target detection, a multi-scale gated attention for adaptive weight allocation (MGA) is put forward, which highlights features of small-target garbage and decreases missed detection probability. The experimental results showed that compared with the benchmark YOLOv7, the detection accuracy in the form of the mean Average Precision (mAP) of APM-YOLOv7 was improved by 7.02%, that of mmAP (mAP0.5:0.95) was improved by 3.91%, and Recall was improved by 11.82%, all of which meet the requirements of high-precision and real-time water-floating garbage detection and provide reliable reference for the intelligent management of water-floating garbage.

Improved UAV-to-Ground Multi-Target Tracking Algorithm Based on StrongSORT.

Unmanned aerial vehicles (UAV) are essential for aerial reconnaissance and monitoring. One of the greatest challenges facing UAVs is vision-based multi-target tracking. Multi-target tracking algorithms that depend on visual data are utilized in a variety of fields. In this study, we present a comprehensive framework for real-time tracking of ground robots in forest and grassland environments. This framework utilizes the YOLOv5n detection algorithm and a multi-target tracking algorithm for monitoring ground robot activities in real-time video streams. We optimized both detection and re-identification networks to enhance real-time target detection. The StrongSORT tracking algorithm was selected carefully to alleviate the loss of tracked objects due to factors like camera jitter, intersecting and overlapping targets, and smaller target sizes. The YOLOv5n algorithm was used to train the dataset, and the StrongSORT tracking algorithm incorporated the best-trained model weights. The algorithm's performance has greatly improved, as demonstrated by experimental results. The number of ID switches (IDSW) has decreased by sixfold, IDF1 has increased by 7.93%, and false positives (FP) have decreased by 30.28%. Additionally, the tracking speed has reached 38 frames per second. These findings validate our algorithm's ability to fulfill real-time tracking requisites on UAV platforms, delivering dependable resolutions for dynamic multi-target tracking on land.

YOLO-FR: A YOLOv5 Infrared Small Target Detection Algorithm Based on Feature Reassembly Sampling Method.

The loss of infrared dim-small target features in the network sampling process is a major factor affecting its detection accuracy. In order to reduce this loss, this paper proposes YOLO-FR, a YOLOv5 infrared dim-small target detection model, based on feature reassembly sampling, which refers to scaling the feature map size without increasing or decreasing the current amount of feature information. In this algorithm, an STD Block is designed to reduce the loss of features during down-sampling by saving spatial information to the channel dimension, and the CARAFE operator, which increases the feature map size without changing the feature mapping mean, is adopted to ensure that features are not distorted by relational scaling. In addition, in order to make full use of the detailed features extracted by the backbone network, the neck network is improved in this study so that the feature extracted after one down-sampling of the backbone network is fused with the top-level semantic information by the neck network to obtain the target detection head with a small receptive field. The experimental results show that the YOLO-FR model proposed in this paper achieved 97.4% on mAP50, which is a 7.4% improvement compared to the original network, and it also outperformed J-MSF and YOLO-SASE.

An Identification Method for Irregular Components Related to Terminal Blocks in Equipment Cabinet of Power Substation.

Despite the continuous advancement of intelligent power substations, the terminal block components within equipment cabinet inspection work still often require loads of personnel. The repetitive documentary works not only lack efficiency but are also susceptible to inaccuracies introduced by substation personnel. To resolve the problem of lengthy, time-consuming inspections, a terminal block component detection and identification method is presented in this paper. The identification method is a multi-stage system that incorporates a streamlined version of You Only Look Once version 7 (YOLOv7), a fusion of YOLOv7 and differential binarization (DB), and the utilization of PaddleOCR. Firstly, the YOLOv7 Area-Oriented (YOLOv7-AO) model is developed to precisely locate the complete region of terminal blocks within substation scene images. The compact area extraction model rapidly cuts out the valid proportion of the input image. Furthermore, the DB segmentation head is integrated into the YOLOv7 model to effectively handle the densely arranged, irregularly shaped block components. To detect all the components within a target electrical cabinet of substation equipment, the YOLOv7 model with a differential binarization attention head (YOLOv7-DBAH) is proposed, integrating spatial and channel attention mechanisms. Finally, a general OCR algorithm is applied to the cropped-out instances after image distortion to match and record the component's identity information. The experimental results show that the YOLOv7-AO model reaches high detection accuracy with good portability, gaining 4.45 times faster running speed. Moreover, the terminal block component detection results show that the YOLOv7-DBAH model achieves the highest evaluation metrics, increasing the F1-score from 0.83 to 0.89 and boosting the precision to over 0.91. The proposed method achieves the goal of terminal block component identification and can be applied in practical situations.

An Infrared Small Target Detection Method Based on Attention Mechanism.

The human visual attention system plays an important role in infrared target recognition because it can quickly and accurately recognize infrared small targets and has good scene adaptability. This paper proposes an infrared small target detection method based on an attention mechanism, which consists of three modules: a bottom-up passive attention module, a top-down active attention module, and decision feedback equalization. In the top-down active attention module, given the Gaussian characteristics of infrared small targets, the idea of combining knowledge-experience Gaussian shape features is applied to implement feature extraction, and quaternion cosine transform is performed to achieve multi-dimensional fusion of Gaussian shape features, thereby achieving complementary fusion of multi-dimensional feature information. In the bottom-up passive attention module, considering that the difference in contrast and motion between the target and the background can attract attention easily, an optimal fast local contrast algorithm and improved circular pipeline filtering are adopted to find candidate target regions. Meanwhile, the multi-scale Laplacian of the Gaussian filter is adopted to estimate the optimal size of the infrared small target. The fast local contrast algorithm based on box filter acceleration and structure optimization is employed to extract local contrast features, and candidate target regions can be obtained by using an adaptive threshold. Besides, the mean gray, target size, Gaussian consistency, and circular region constraint are used in pipeline filtering to extract motion regions, and the false-alarm rate is reduced effectively. Finally, decision feedback equalization is adopted to obtain real targets. Experiments are conducted on some real infrared images involving complex backgrounds with sea, sky, and ground clutters, and the experimental results indicate that the proposed method can achieve better detection performance than conventional baseline methods, such as RLCM, ILCM, PQFT, MPCM, and ADMD. Also, mathematical proofs are provided to validate the proposed method.

Enhancing UAV Detection in Surveillance Camera Videos through Spatiotemporal Information and Optical Flow.

The growing intelligence and prevalence of drones have led to an increase in their disorderly and illicit usage, posing substantial risks to aviation and public safety. This paper focuses on addressing the issue of drone detection through surveillance cameras. Drone targets in images possess distinctive characteristics, including small size, weak energy, low contrast, and limited and varying features, rendering precise detection a challenging task. To overcome these challenges, we propose a novel detection method that extends the input of YOLOv5s to a continuous sequence of images and inter-frame optical flow, emulating the visual mechanisms employed by humans. By incorporating the image sequence as input, our model can leverage both temporal and spatial information, extracting more features of small and weak targets through the integration of spatiotemporal data. This integration augments the accuracy and robustness of drone detection. Furthermore, the inclusion of optical flow enables the model to directly perceive the motion information of drone targets across consecutive frames, enhancing its ability to extract and utilize features from dynamic objects. Comparative experiments demonstrate that our proposed method of extended input significantly enhances the network's capability to detect small moving targets, showcasing competitive performance in terms of accuracy and speed. Specifically, our method achieves a final average precision of 86.87%, representing a noteworthy 11.49% improvement over the baseline, and the speed remains above 30 frames per second. Additionally, our approach is adaptable to other detection models with different backbones, providing valuable insights for domains such as Urban Air Mobility and autonomous driving.

Micro-Leakage Image Recognition Method for Internal Detection in Small, Buried Gas Pipelines.

In order to resolve the problem that the sample of image for internal detection of DN100 buried gas pipeline microleakage is single and difficult to identify, a recognition method of microleakage image of the pipeline internal detection robot is proposed. First, nongenerative data augmentation is used to expand the microleakage images of gas pipelines. Secondly, a generative data augmentation network, Deep Convolutional Wasserstein Generative Adversarial Networks (DCWGANs), is designed to generate microleakage images with different features for detection in the pipeline of gas pipelines to achieve sample diversity of microleakage images of gas pipelines. Then, a bi-directional feature pyramid network (BiFPN) is introduced into You Only Look Once (YOLOv5) to retain more deep feature information by adding cross-scale connecting lines in the feature fusion structure; finally, a small target detection layer is constructed in YOLOv5 so that more shallow feature information can be retained to achieve small-scale leak point recognition. The experimental results show that the precision of this method for microleak identification is 95.04%, the recall rate is 94.86%, the mAP value is 96.31%, and the minimum size of identifiable leaks is 1 mm.

ST-CenterNet: Small Target Detection Algorithm with Adaptive Data Enhancement.

General target detection with deep learning has made tremendous strides in the past few years. However, small target detection sometimes is associated with insufficient sample size and difficulty in extracting complete feature information. For safety during autonomous driving, remote signs and pedestrians need to be detected from driving scenes photographed by car cameras. In the early period of a medical lesion, because of the small area of the lesion, target detection is of great significance to detect masses and tumors for accurate diagnosis and treatment. To deal with these problems, we propose a novel deep learning model, named CenterNet for small targets (ST-CenterNet). First of all, due to the lack of visual information on small targets in the dataset, we extracted less discriminative features. To overcome this shortcoming, the proposed selective small target replication algorithm (SSTRA) was used to realize increasing numbers of small targets by selectively oversampling them. In addition, the difficulty of extracting shallow semantic information for small targets results in incomplete target feature information. Consequently, we developed a target adaptation feature extraction module (TAFEM), which was used to conduct bottom-up and top-down bidirectional feature extraction by combining ResNet with the adaptive feature pyramid network (AFPN). The improved new network model, AFPN, was added to solve the problem of the original feature extraction module, which can only extract the last layer of the feature information. The experimental results demonstrate that the proposed method can accurately detect the small-scale image of distributed targets and simultaneously, at the pixel level, classify whether a subject is wearing a safety helmet. Compared with the detection effect of the original algorithm on the safety helmet wearing dataset (SHWD), we achieved mean average precision (mAP) of 89.06% and frames per second (FPS) of 28.96, an improvement of 18.08% mAP over the previous method.

A Combined Approach to Infrared Small-Target Detection with the Alternating Direction Method of Multipliers and an Improved Top-Hat Transformation.

In infrared small target detection, the infrared patch image (IPI)-model-based methods produce better results than other popular approaches (such as max-mean, top-hat, and human visual system) but in some extreme cases it suffers from long processing times and inconsistent performance. In order to overcome these issues, we propose a novel approach of dividing the traditional target detection process into two steps: suppression of background noise and elimination of clutter. The workflow consists of four steps: after importing the images, the second step applies the alternating direction multiplier method to preliminarily remove the background. Comparatively to the IPI model, this step does not require sliding patches, resulting in a significant reduction in processing time. To eliminate residual noise and clutter, the interim results from morphological filtering are then processed in step 3 through an improved new top-hat transformation, using a threefold structuring element. The final step is thresholding segmentation, which uses an adaptive threshold algorithm. Compared with IPI and the new top-hat methods, as well as some other widely used methods, our approach was able to detect infrared targets more efficiently (90% less computational time) and consistently (no sudden performance drop).

Infrared Single-Frame Small Target Detection Based on Block-Matching.

The robust detection of small targets is one of the crucial techniques in an infrared system. It is still a challenge to detect small targets under complex backgrounds. Aiming at the problem where infrared small target detection is easily disturbed by complex backgrounds, an infrared single frame detection method based on a block-matching approach is proposed in this paper. Firstly, the input infrared image is processed by extracting blocks from it. A new infrared model is constructed by finding blocks that are similar to each such block. Then, the small target detection based on the block-matching model is formulated as an optimization problem of recovering low-rank and sparse matrices, which are effectively solved using robust principal component analysis. Finally, the results of processing are reconstructed to obtain the target and background images. A simple segmentation method is used to segment the target image. The experimental results from the actual infrared sequences show that the proposed method has better background suppression ability under complex backgrounds and better detection performance than conventional baseline methods.

Helmet Wearing State Detection Based on Improved Yolov5s.

At many construction sites, whether to wear a helmet is directly related to the safety of the workers. Therefore, the detection of helmet use has become a crucial monitoring tool for construction safety. However, most of the current helmet wearing detection algorithms are only dedicated to distinguishing pedestrians who wear helmets from those who do not. In order to further enrich the detection in construction scenes, this paper builds a dataset with six cases: not wearing a helmet, wearing a helmet, just wearing a hat, having a helmet, but not wearing it, wearing a helmet correctly, and wearing a helmet without wearing the chin strap. On this basis, this paper proposes a practical algorithm for detecting helmet wearing states based on the improved YOLOv5s algorithm. Firstly, according to the characteristics of the label of the dataset constructed by us, the K-means method is used to redesign the size of the prior box and match it to the corresponding feature layer to increase the accuracy of the feature extraction of the model; secondly, an additional layer is added to the algorithm to improve the ability of the model to recognize small targets; finally, the attention mechanism is introduced in the algorithm, and the CIOU_Loss function in the YOLOv5 method is replaced by the EIOU_Loss function. The experimental results indicate that the improved algorithm is more accurate than the original YOLOv5s algorithm. In addition, the finer classification also significantly enhances the detection performance of the model.

On-Machine Detection of Sub-Microscale Defects in Diamond Tool Grinding during the Manufacturing Process Based on DToolnet.

Nowadays, tool condition monitoring (TCM), which can prevent the waste of resources and improve efficiency in the process of machining parts, has developed many mature methods. However, TCM during the production of cutting tools is less studied and has different properties. The scale of the defects in the tool production process is tiny, generally between 10 µm and 100 µm for diamond tools. There are also very few samples with defects produced by the diamond tool grinding process, with only about 600 pictures. Among the many TCM methods, the direct inspection method using machine vision has the advantage of obtaining diamond tool information on-machine at a low cost and with high efficiency, and the method is accurate enough to meet the requirements of this task. Considering the specific, above problems, to analyze the images acquired by the vision system, a neural network model that is suitable for defect detection in diamond tool grinding is proposed, which is named DToolnet. DToolnet is developed by extracting and learning from the small-sample diamond tool features to intuitively and quickly detect defects in their production. The improvement of the feature extraction network, the optimization of the target recognition network, and the adjustment of the parameters during the network training process are performed in DToolnet. The imaging system and related mechanical structures for TCM are also constructed. A series of validation experiments is carried out and the experiment results show that DToolnet can achieve an 89.3 average precision (AP) for the detection of diamond tool defects, which significantly outperforms other classical network models. Lastly, the DToolnet parameters are optimized, improving the accuracy by 4.7%. This research work offers a very feasible and valuable way to achieve TCM in the manufacturing process.

YOLO-SASE: An Improved YOLO Algorithm for the Small Targets Detection in Complex Backgrounds.

To improve the detection ability of infrared small targets in complex backgrounds, an improved detection algorithm YOLO-SASE is proposed in this paper. The algorithm is based on the YOLO detection framework and SRGAN network, taking super-resolution reconstructed images as input, combined with the SASE module, SPP module, and multi-level receptive field structure while adjusting the number of detection output layers through exploring feature weight to improve feature utilization efficiency. Compared with the original model, the accuracy and recall rate of the algorithm proposed in this paper were improved by 2% and 3%, respectively, in the experiment, and the stability of the results was significantly improved in the training process.