163results about How to "Reduce false detection rate" patented technology

The invention discloses a distributed photovoltaic panel anomaly detection method under a high-altitude view angle based on a YOLOv8n model, and belongs to the technical field of target detection. The model comprises the following steps: acquiring a distributed photovoltaic panel image data set under a high-altitude view angle, and performing data enhancement and labeling; an improved YOLOv8n model is constructed, and the improvement comprises the steps that a C2fAT module is introduced into a backbone network, and the small target feature extraction capacity is enhanced; an SPPF module is replaced by an SPPF-LSKA module, and complex background interference is suppressed; an EMA attention mechanism is introduced into the neck network, and the multi-scale adaptive capacity is improved; optimizing a training process by adopting a WIoU v3 loss function; and training and optimizing the model by using the training set, and finally outputting a detection result through the test set. According to the invention, the problems of false detection and missing detection of the distributed photovoltaic panel in a high-altitude view angle are effectively solved, the detection precision of the distributed photovoltaic panel is further improved, and inspection personnel are helped to troubleshoot the photovoltaic panel in an abnormal state in the high-altitude view angle.

The invention discloses a life body feature separation method based on millimeter-wave radar signal enhancement, and the method comprises the following steps: obtaining an original radar signal, collected by a millimeter-wave radar, of a life body monitoring target, and carrying out the preprocessing of the original radar signal; six types of features are extracted from the preprocessed signals, and the preprocessed signals are divided into four scene types based on the six types of features; according to the divided scene categories, calling a corresponding signal processing model to enhance or reconstruct the preprocessed signal to obtain an enhanced radar signal; and performing signal quality evaluation on the enhanced radar signal, inputting the enhanced signal meeting the quality evaluation requirement into a variational mode decomposition module, and separating a respiration signal and a heartbeat signal of the life body monitoring target. According to the method, the millimeter wave radar vital sign signals in a complex scene are subjected to scene classification and adaptive calling of the model to enhance the signals, breathing and heartbeat are separated through variational mode decomposition, and the vital sign detection precision in a complex environment is improved.

The application discloses a fan shell front and rear hole position positioning method and system based on image recognition, comprising the following steps: S1: collecting a fan shell image through a calibrated binocular camera, and performing stereo correction and illumination invariance enhancement processing to obtain binocular enhanced image data; S2: performing instance segmentation on the binocular enhanced image data based on an instance segmentation neural network to obtain a hole position pixel region set with front and rear semantic labels; S3: performing data-driven hole position coarse matching on the hole position pixel region set through a graph neural network, and constructing a physical constraint rule model to optimize and verify the matching result, and outputting an accurate image matching pair set of the front and rear hole positions; and S4: based on the accurate image matching pair set, constructing a cost function to solve the shell pose and obtain final world coordinates, and realizing the front and rear hole position positioning of the fan shell.

This invention discloses a system and method for labeling automotive abnormal noise data, relating to the field of automotive NVH detection technology. The method includes: receiving raw signal data of automotive abnormal noise; performing time-frequency transformation on the raw signal to generate a time-frequency diagram; identifying key frequency ranges characterizing the abnormal noise features; then performing directional filtering to enhance the abnormal noise signal; identifying candidate time ranges for abnormal noise through an automatic detection algorithm; and outputting the final abnormal noise time range labels and key frequency ranges in a structured format after verification and correction. The system includes a data receiving module, a signal processing module, a display module, an intelligent labeling module, an audio playback module, a labeling output module, and an interaction module, achieving triple-assisted labeling through auditory, visual, and intelligent pre-selection methods. This solves the problems of low efficiency, low accuracy, and poor consistency in traditional manual listening methods, providing high-quality labeled data for deep learning classification of automotive abnormal noises.

The present application belongs to the field of defect detection, in particular to a pipe network weld flaw detection defect recognition method and system, comprising: acquiring a pipe network weld initial radiographic image, extracting a global feature atlas through a first deep convolution network, and decoding to generate a defect prediction confidence map; extracting a weld key geometric structure, and constructing a geometric prior weight map; applying Bayesian variational inference to the network, statistically dispersing the results of multiple random forward propagation, representing cognitive uncertainty and generating an uncertainty map; pixel-level weighted fusion of the confidence map, the uncertainty map and the geometric prior weight map to obtain a probability heat map, based on which a composite sampling guide vector field is constructed, sampling points are arranged along the vector field, and a multi-angle scanning imaging system is controlled to collect high-resolution local projection data; three-dimensional reconstruction of the projection data to obtain local features, fusion of the local features and global features through a cross-attention module to generate an enhanced defect representation, and output of the class, three-dimensional spatial position and size of the weld defect.

The present application belongs to the field of power system intelligent inspection and computer vision technology, and provides a kind of transmission line hidden danger target detection test adaptive method and system, obtains inspection image, utilizes pre-training detector to carry out forward inference to the inspection image, generates original prediction set, and the original prediction set includes prediction frame coordinates, class label and confidence score;Geometric verification is carried out, and reasonable geometric prediction set is obtained;Carry out spatial neighborhood clustering, obtain multiple candidate clusters, calculate the scale dispersion of the area of the prediction frame in each candidate cluster, group according to the class label, dynamically cluster the prediction frame in the group, perform weighted average on the prediction frame coordinates with the confidence score as the weight, calculate the fusion confidence, obtain the refined pseudo label set, use it as the supervision signal, calculate the detection loss, and iteratively update the model parameters through the optimization algorithm until the iteration number is satisfied.The present application improves accuracy, versatility and engineering reliability.

This invention belongs to the field of intelligent detection technology for wind turbine blades, specifically relating to a feature detection method and system for multi-category surface damage on blades. The method includes: constructing a pyramid network structure EFPN to replace the baseline model's FPN+PAN neck network. The EFPN includes a pyramid feature enhancement (PFE) module, a multi-feature fusion (MFF) module, and a dynamic interpolation fusion (DIF) module, used to integrate multi-scale damage features and suppress background noise; designing a Focaler PIoU loss function, combined with a size-adaptive penalty factor and a dynamic sample weighting mechanism, to improve the model's ability to perceive difficult samples and the accuracy of bounding box regression; and using the DepGraph structured pruning method to compress the network, combined with an HDCFR hybrid knowledge distillation strategy, to reduce model complexity while maintaining detection accuracy. This invention achieves an mAP@0.5 of 84.5% on the WTBs GLE dataset, with reduced parameter and computational costs, combining high accuracy, lightweight design, and strong robustness, making it suitable for deployment on UAV edge devices.

This invention belongs to the technical field of aviation transponder pulse overlap signal separation methods, specifically relating to a separation method applicable to Mode A / C aviation transponder pulse collision conditions. By analyzing the differential characteristics of the Mode A / C aviation transponder signal waveform in the time and energy domains, the pulse collision signal is effectively split. Based on the waveform definition characteristics of the Mode A / C aviation transponder signal, the pulse frame is accurately recovered from the separated signal, ultimately completing the separation process of the Mode A / C collision pulse signal. This invention requires only a single-channel receiver to achieve signal separation, reducing system implementation costs and engineering difficulty, effectively improving the system monitoring capability in densely overlapping electromagnetic environments, and providing an efficient and engineered solution for transponder pulse signal separation tasks in scenarios such as air traffic surveillance and low-altitude airspace management.

The invention discloses a honeycomb sandwich structure infrared image defect detection method and system based on improved YOLOv5. The method comprises the following steps: collecting defect images of a honeycomb sandwich structure to form a sample data set; training defect images in the data set based on an improved YOLOv5 model, predicting positions and categories of honeycomb sandwich structure defects, training an optimization model by using a loss function until convergence, and obtaining and storing optimal model weight data; the improved YOLOv5 model is based on a YOLOv5 architecture, a Swin Transform module is embedded in a backbone network, and the detection performance of the model on honeycomb sandwich structure defects is improved in combination with a C3 convolutional layer, a CBS module and an SPPF module; and acquiring a defect image of the honeycomb sandwich structure to be detected, detecting and identifying the defect image by using the improved YOLOv5 model based on the optimal model weight data, and generating and outputting a detection result. According to the method, small-size defect identification in multi-scale defect identification can be effectively realized, and the defects of a traditional convolutional neural network in processing complex backgrounds and small target detection are overcome.

This invention provides a method and apparatus for vehicle scratch recognition based on a cascaded semantic segmentation and target detection, belonging to the field of computer vision technology. The method includes: acquiring a vehicle exterior image to be detected; performing semantic segmentation on the vehicle exterior image to obtain a corresponding vehicle foreground probability map; binarizing the vehicle foreground probability map to obtain a binary mask; performing background removal processing on the binary mask to obtain a background-removed image corresponding to the vehicle exterior image; and calling a scratch detection model to perform scratch recognition on the background-removed image to obtain the vehicle scratch recognition result. This invention addresses the technical problem of high false detection rate in existing vehicle surface scratch recognition technologies.

This invention belongs to the field of computer vision and image processing technology, and particularly relates to a query key-guided asymmetric feature enhancement system and method. The system includes: Q-ASFE modules embedded in the backbone network, specifically in the link where features are transmitted from the C3K2 module to the CBS module. The P2 and P3 layer feature maps output by the backbone network are processed by the neck network to output three feature fusion maps at different scales. These three feature fusion maps are then processed by a detection head to obtain the final detection result. This invention achieves efficient enhancement of image features through the organic integration of spatial detail awareness and global semantic guidance.

The application belongs to the technical field of vehicle information anonymization detection, and particularly relates to a multi-type image anonymization annotation dataset construction and target coverage rate determination method. The method is based on a face and license plate image dataset with double annotation of theoretical anonymization region and anonymization features, introduces an anonymization feature extraction branch in an improved YOLOv5-L model and performs cross-modal feature fusion, combines a plurality of loss functions with theoretical anonymization region positioning loss as the core and a two-stage progressive training and difficult example mining mechanism, and realizes precise learning of the anonymization features. After normalizing the input anonymization image, the model inference obtains the theoretical anonymization region coordinates and maps them back to the original size, and through non-maximum suppression and matching of the IoU threshold, the region coverage rate is calculated to determine the positive detection, missed detection and statistical false detection rate. The application effectively overcomes the feature dependency failure and model robustness problem, and realizes high-precision, low-misjudgment anonymization detection and evaluation under various anonymization conditions.

The utility model discloses a desktop formula wire joint piece scattered silk defect detection device, including the shell body that is located on the desktop, the bottom inner wall of shell body is installed with high definition imaging mechanism, and the front panel on shell body is installed with detection mechanism, the high definition imaging mechanism includes the inclined base fixed mounting on the bottom inner wall of shell body, and the top inclined surface of inclined base is provided with shooting assembly and adjusting assembly, the detection mechanism includes the plug board of inlay formula fixed mounting on the front panel of shell body, and the inboard of plug board is fixedly connected with glass plate through connecting column, and the side fixed mounting of glass plate near plug board has multiple prisms that are circularly distributed, the utility model discloses through high resolution's industrial camera and advanced image recognition algorithm, can accurately identify the tiny defect of wire joint piece, effectively reduces the misjudgment rate, improves product quality, and simple operation simultaneously, and the design structure of desktop formula is compact, and the floor space is small, and is convenient for installation and operation.

This invention provides a method for detecting rotating debris in coal flow based on physical fuzzy modeling and scene-customized attention. The method includes: acquiring a clear coal flow image and then labeling non-coal debris targets in the image; establishing a dataset; constructing a debris detection network; constructing a physics-driven motion fuzzy mapping model; using the motion fuzzy mapping model to dynamically blur the clear coal flow image, generating training samples simulating motion fuzziness phenomena in a conveyor belt; training and optimizing the debris detection network; inputting a real-time coal flow image into the optimized network to obtain candidate detection results for non-coal debris targets; post-processing the candidate detection results and outputting the final filtered detection results. This invention can improve the stability and accuracy of debris detection under complex conditions of high-speed coal flow, reduce the influence of motion fuzziness and background interference, and provide technical support for the safety monitoring and automatic sorting of coal mine conveying systems.

This invention belongs to the field of medical image processing technology, specifically relating to a method for filtering and enhancing endoscopic video images. This invention achieves reliable restoration of tissue texture beneath specular highlight-occluded areas in endoscopic videos through five processing stages. The first stage uses a three-factor joint decision to adaptively detect highlight regions in video frames and generate a time-varying highlight mask, effectively distinguishing between moving highlights and fixed bright spots. The second stage uses hierarchical sparse Lucas-Kanade optical flow tracing and RANSAC affine transformation estimation to accurately map highlight-occluded pixels to corresponding non-highlight tissue locations in historical frames, constructing a candidate texture set. The third stage uses a two-color reflectance model to extract tissue intrinsic colors and perform illumination normalization migration on the candidate textures, eliminating inter-frame illumination differences. The fourth stage constructs confidence scores using the product of three-dimensional weights and obtains the restored color through weighted median aggregation. The fifth stage applies guided filtering and inter-frame exponential smoothing to output a temporally consistent enhanced video.

The application discloses an online detection system and method for a digital printing film switch, and relates to the technical field of machine vision detection. The method comprises the following steps: acquiring an initial image of a product to be detected; processing the initial image to obtain a second image; detecting the second image to obtain a size defect result, a position defect result, a color difference defect result, a bubble defect result and a crack defect result of the product to be detected; obtaining a total defect detection result of the product to be detected according to the size defect result, the position defect result, the color difference defect result, the bubble defect result and the crack defect result; and sorting the product to be detected according to the total defect detection result of the product to be detected. The method improves detection precision and production efficiency, and effectively reduces the missing detection rate and the false detection rate.

This application discloses a security protection method and related products. The method includes: identifying abnormal behavior in the data to be processed corresponding to a target vehicle based on a preset security rule base and a pre-trained anomaly recognition model to obtain a security recognition result; if the security recognition result indicates a first anomaly level, then performing local security protection on the target vehicle according to a local protection strategy; if the security recognition result indicates a second anomaly level, then obtaining the collaborative protection strategy sent by the cloud platform corresponding to the target vehicle, and performing collaborative security protection with other vehicles according to the collaborative protection strategy. This improves the security of the entire vehicle-to-everything (V2X) environment.

The application provides a UAV remote sensing image target detection method and device, and relates to the technical field of UAV target detection.The lightweight main network with GhostNetV2 as the core is used to compress the model parameter quantity and the calculation cost, and meanwhile, the shallow detail features of small targets are completely reserved, thereby laying a stable feature foundation for small target detection; through the dynamic sparse feature fusion and context enhancement operation of the lightweight dynamic feature pyramid network, the fusion of small target sensitive feature paths is adaptively strengthened, the small target feature loss problem caused by the fixed fusion strategy is solved, and the small target missing detection rate is significantly reduced; through the double attention enhancement and decoupling detection operation of the lightweight bidirectional attention detection head, the complex background noise is inhibited, the small target feature saliency is strengthened, the optimization conflict of the classification and regression tasks is avoided, the false detection rate is reduced, and the detection precision of the small target of the UAV remote sensing image is improved.

The application relates to the technical field of flaw detection, and discloses a flaw detection device for numerical control machine tool part machining, which comprises a conveying frame with supporting legs, a conveying mechanism is arranged in the conveying frame, the upper surface of one end of the conveying frame is fixedly connected with a detection box, entrances and exits are arranged on the two sides of the detection box close to the chain plate, the middle part of the upper surface of the detection box is fixedly connected with a mounting rod, the lower end of the mounting rod penetrates into the inside of the detection box and is fixedly connected with a mounting block, the bottom surface of the mounting block is fixedly connected with a visual detection camera, one side of the detection box close to the conveying frame is connected with a moving mechanism, and the output end of the moving mechanism is connected with an L-shaped plate. The flaw detection device for numerical control machine tool part machining is provided with the L-shaped plate carrying a standard part, can automatically move to the position below the camera in the detection gap to complete self-checking and calibration, does not need to stop and maintain, and solves the problem of efficiency loss caused by traditional regular stop and calibration.

The invention provides a red deer identification and detection method, system and device and a medium, and belongs to the technical field of computer vision. The method comprises the following steps: acquiring an ortho-image of a red deer activity area through an unmanned aerial vehicle, and processing the ortho-image by using an illumination adaptive enhancement network to obtain an enhanced image; annotating the enhanced image, and constructing a data set; an improved YOLOv8 deep learning model is constructed and trained, an attention mechanism module fused by GAM and CBAM is inserted in a multi-scale feature fusion path between a feature fusion network and a detection head and in the detection head, and four detection heads with different scales are included. Wherein the tiny, small-scale and medium-scale detection heads are integrated with sub-branches for predicting biological attributes; in the training process, the bounding box, the category and the attribute tag are utilized to calculate the composite loss for optimization; and finally evaluating the model performance by using the test set. The method can effectively improve the detection precision and robustness of red deer target recognition in a complex natural scene.

The invention relates to the technical field of agricultural remote sensing monitoring, in particular to a method and system for measuring the number (area) of plants in a tobacco field based on visible light images of an unmanned aerial vehicle, and is suitable for automatic and high-precision growth monitoring and resource accounting of a large-scale tobacco field. Comprising six steps of tobacco field visible light image acquisition, image preprocessing, tobacco field region segmentation, tobacco field plant target detection and plant number statistics, tobacco field area calculation, and result post-processing and output, and through combination of an improved target detection algorithm and an image segmentation technology, high precision of tobacco field plant number statistics and high accuracy of area determination are realized. Meanwhile, the detection efficiency is guaranteed, the actual requirement of large-scale tobacco field monitoring is met, data acquisition, processing, analysis and output can be automatically completed in the whole process, manual intervention is not needed, the monitoring efficiency is greatly improved, and the labor cost is reduced.

The invention discloses a device and a method for detecting tiny flaws on the surface of a reflective vehicle logo, and relates to the technical field of defect detection. The method comprises the following steps: acquiring camera acquisition data, and preprocessing the camera data to obtain a repaired image; a cross-space multi-scale attention module is added in the YOLOv8 network, and a defect detection model comprising a target scale mapping module and the cross-space multi-scale attention module is formed; inputting the repaired image into a flaw detection model for training to obtain a trained flaw detection model; and inputting a to-be-detected image into the trained flaw detection model, and detecting tiny flaws on the surface of the reflective vehicle logo. According to the invention, reflection interference on the surface of the vehicle logo can be effectively inhibited by using the polarization technology, the image acquisition quality is improved, the method is suitable for vehicle logo of different specifications, and a reliable basis is provided for subsequent high-precision online real-time defect identification.

The present application belongs to the technical field of remote sensing image target detection, and particularly relates to a remote sensing image open vocabulary target detection method. The present application realizes open vocabulary, knowledge-enhanced and interpretable target detection of remote sensing images by constructing a knowledge graph that can be incrementally expanded online and deeply cooperating the knowledge graph with a visual language model. The present application can understand and respond to any new category input by a user in real time without retraining, greatly improving the application flexibility. The present application simultaneously applies the prior information of the knowledge graph to two key links. At the task level, the task instruction text fused with the prior information of the knowledge graph is encoded into a task condition semantic vector, and the parameters of the visual language model are adjusted according to the task condition semantic vector. At the concept level, the prior information of the knowledge graph is input into the visual language model to construct a stable semantic embedding vector for the new category. The two cooperate to effectively solve the semantic drift and adaptation problems of general models in the remote sensing field.

The application discloses a kind of dynamic rotating heliostat mirror damage identification method based on unmanned aerial vehicle;The core is to build the intelligent perception architecture of "air-ground cooperation", the video stream of moving mirror is collected in real time by visual sensor carried on unmanned aerial vehicle platform;Dynamic image stabilization algorithm based on depth feature optical flow and motion compensation is used to offset the coupling interference of unmanned aerial vehicle pose disturbance and target motion;At the same time, a kind of lightweight, attention-guided multiscale attention fusion network is designed, which realizes pixel-level sensitivity and millimeter-level identification of typical defects such as microcracks, stains and coating peeling;Finally, through the spatio-temporal correlation reasoning engine, the damage quantification atlas with geographic information label and operation and maintenance decision suggestion are output.The application breaks through the technical limitations of static or slow target detection, realizes the full-automatic, high-frequency and high-reliable health state inspection of large-scale rotating mirror array, and significantly improves the intelligent operation and maintenance level and power generation efficiency guarantee capability of the solar thermal power station.

This application relates to the field of copper rod defect detection technology, and particularly to an online method and related apparatus for determining subcutaneous porosity in copper rod castings. The method includes: acquiring first detection parameters and a transmission speed; a first detection device performing a first frequency sweep operation on a continuously moving copper rod casting based on the first detection parameters to acquire first detection information in real time; determining second detection parameters of a second detection device in real time based on the first detection information and the transmission speed; the second detection device performing a second frequency sweep operation on the continuously moving copper rod casting based on the second detection parameters to acquire second detection information in real time; and determining porosity defect information based on the first detection information, the second detection information, and the transmission speed; wherein, porosity defect information refers to the location of subcutaneous pores in the copper rod casting. The online method for determining subcutaneous porosity in copper rod castings provided by this application can improve the detection accuracy of subcutaneous porosity defects in copper rod castings.

The invention discloses a sonar point cloud abnormal value detection method for a self-adaptive multi-scale attention mechanism, and the method comprises the steps: collecting original sonar data, carrying out the preprocessing, and generating standardized sonar point cloud data; constructing an acoustic propagation field tensor, calculating propagation direction characteristics, and forming a multi-scale coherence domain; the propagation direction features act on all coherence domains, cross-scale thrust flow response calculation is executed, and propagation evolution data are formed; respectively constructing three branches by taking propagation consistency representation as input to obtain three types of depth prediction results; according to the three types of depth prediction results and the original depth, three types of consistency deviations are constructed, and an abnormal score is generated; and setting an abnormal point judgment threshold, outputting an abnormal point detection result, and eliminating abnormal points. According to the method, the acoustic propagation field tensor is constructed, and multi-scale attention and three-branch consistency reasoning are combined, so that high-precision and self-adaptive detection of the complex abnormal points in the sonar point cloud is realized.

This invention provides a visual detection method and system for coating uniformity, belonging to the field of image processing. The method includes: obtaining the corresponding pixels in each modal image based on the positions of pixels in four matching pairs among all matching pairs of the brightfield image of the coating to be detected and the brightfield image corresponding to each modality; dividing the brightfield image of the coating to be detected into several regions based on the value of each pixel in the H channel of the HSV color space; and obtaining the uneven coating region based on the grayscale value of the corresponding pixel in the darkfield image and the relative thickness of the corresponding pixel in the narrowband spectral image within each region of the brightfield image of the coating to be detected. This invention aims to solve the problem of relatively low accuracy in detecting coating uniformity using a single modal image.

This invention discloses a method and system for detecting short circuits or open circuits in silicon or glass wafer-based solar panels based on multimodal sensing collaboration, belonging to the field of photovoltaic testing technology. The system includes a testing platform structure, a pattern generator, a probe structure, a probe platform structure, a detection module, a platform structure, and a data processing module. The detection module can utilize a capacitance sensor, a magnetic field sensor (parallel / vertical magnetic field components), an IR camera, a UV / visible light camera, or a resistance measurement unit. By applying AC / DC / mixed signals to the electrodes, it acquires voltage distribution, magnetic field components, heat generation, or optical images. The data processing module analyzes and locates the short circuit or open circuit location. This invention integrates multiple detection modes, achieving a positioning accuracy of ±5µm, supporting multi-source data fusion, adapting to the detection of silicon / glass wafer-based solar panels of different sizes, and significantly improving detection efficiency and accuracy to meet the needs of industrial production lines.

The application discloses a kind of modular configurable tooling detection methods, belong to tooling detection field, specifically include: module configuration step, automatic detection step, detection result processing step.Before detection, detection group leader configures detection hardware according to detection plan, software platform.When the unit to be detected is placed in the area to be detected, detection personnel start main control machine, and start automatic detection process according to the configuration parameters of detection scheme module, detection strategy module and detection alarm module after authentication.After detection, whether the detection data is entered into database according to the parameters configured by data analysis module is processed by main control software;Whether detection report is saved or printed;The application is adapted to the detection needs of the unit to be detected constantly upgrading iteration by the modularization and configurability of detection hardware and software platform, and the problem that error reproduction is difficult, error chain is deep and difficult to locate is solved by error tracing module.

This invention discloses a bidirectional guided fusion method for image information, a target detection method, and a system. The fusion method includes: inputting the image information to be processed into a YOLO model to obtain a shallow feature image from the Backbone network and a deep feature image from the Neck network; decoupling the feature frequency bands of the shallow feature image and the upsampled deep feature image to obtain a shallow high-frequency feature map and a deep low-frequency feature map, respectively; analyzing the semantic information represented by the deep low-frequency feature map to guide the shallow high-frequency feature map to enhance details, resulting in a detail-enhanced image; evaluating the weights of each feature channel using the shallow high-frequency feature map to guide the deep low-frequency feature map to calibrate semantics, resulting in a calibrated image; and fusing the detail-enhanced image with the calibrated image. This invention solves the problems of detail loss and semantic ambiguity in existing target detection networks, especially for small-scale targets.