147 results about "Background suppression" patented technology

The invention discloses a remote sensing small target detection network and method based on frequency domain and space domain adaptive enhancement, and belongs to the technical field of artificial intelligence. The network is based on an FCOS lightweight detection framework, through cooperative extraction, adaptive enhancement and fusion of spatial domain and frequency domain features, the parameter quantity and the calculation complexity are greatly reduced, the small target detection precision and efficiency are significantly improved, and finally the balance of lightweight deployment and high-precision detection is realized. According to the invention, through the space-frequency feature adaptive enhancement network, space context modeling and frequency detail recovery can be completed in a single-stage detection framework at the same time, and an adaptive enhancement unit is introduced to dynamically adjust the enhancement strength according to the local signal-to-noise ratio, so that significant improvement and background suppression of a tiny target signal are realized.

The invention discloses a cable sheath microcrack image identification method based on deep learning. The method comprises the following steps: acquiring a cable sheath image and executing image preprocessing operation; inputting to an improved MAE model, and generating a background reconstruction image and a crack reconstruction image; pixel-level residual fusion is carried out to generate a background shielding image; performing pixel-level fusion on the background shielding image and the preprocessed image to generate a background suppression image; micro-crack recognition operation is executed, and a preliminary crack response heat map set is output through image feature extraction and crack region judgment; executing a heat map accumulative analysis operation, and constructing a multi-scale accumulative heat map; judging a pseudo response risk area according to the local response change rate; response value retraction operation is executed based on the corresponding local area, and a crack heat map after pseudo response suppression is generated; and extracting a high-confidence crack region to obtain a cable sheath microcrack identification result. According to the invention, the precision and robustness of microcrack detection are improved, and the background interference and false detection risk are reduced.

The invention relates to the field of computer vision, artificial intelligence and intelligent fitness, and particularly discloses a fitness action recognition, counting and quality evaluation method based on machine vision, which comprises the following steps of: extracting video frames and standardizing the video frames, realizing background suppression and human body region enhancement through a semantic segmentation or inter-frame difference method, and outputting a standardized image sequence; detecting key joint points by adopting a pre-training model, and outputting a stable skeleton sequence and candidate action stage data through time sequence consistency filtering; reconstructing a feature tensor, fusing spatio-temporal features through double-branch attention collaboration, and outputting action categories and time sequence compensation parameters in a classified manner; a dynamic threshold method accumulates the number of actions, action qualification is judged in combination with biomechanical constraints, the confidence coefficient is optimized, and a structured result is output; bone rendering, error highlighting, voice generation and personalized training suggestions. According to the method, wearable equipment is not needed, the problems of instable identification, miscounting and the like in a complex scene are solved, and real-time accurate analysis is realized.

The invention discloses a gas leakage detection imaging device and method based on a single photon detection technology, and belongs to the technical field of gas detection and imaging. Aiming at the limitation problems of the existing gas leakage detection technology in the aspects of sensitivity, response speed, space imaging capability and anti-interference capability, nanosecond modulation control and synchronous photon counting are realized by utilizing an FPGA (Field Programmable Gate Array) through a visual gas detection imaging device combining an SPAD (Single Program Amplifier) array detector and a WMS modulation-demodulation mechanism; and the second harmonic of the absorption signal is further extracted by combining digital phase-locked amplification, so that gas leakage three-dimensional imaging with high spatial resolution, ppb-level sensitivity, strong background suppression capability and high time response is realized. The system has the advantages of high detection sensitivity, real-time imaging capability with high spatial resolution, high background light interference resistance and high signal-to-noise ratio, the demodulation method is novel, and the modular design has multi-gas compatibility and system expansibility.

The invention discloses an AI real-time small target detection method based on attention, and relates to the technical field of computer vision and deep learning. The AI real-time small target detection method based on attention comprises a small target feature enhancement layer, a dynamic background suppression layer, a lightweight detection optimization layer and an edge deployment adaptation layer. The small target feature enhancement layer is used for enhancing small target features and making up feature loss, and the dynamic background suppression layer is used for accurately filtering background interference. According to the AI real-time small target detection method based on attention, through a feature enhancement layer and loss optimization, on a VisDrone2022 data set, small target detection mAP (at) 0.5 reaches 72.3%, the mAP (at) is improved by 18.2% compared with a YOLOv5 basic version (54.1%), and the mAP (at) is improved by 25.5% compared with a Faster R-CNN (46.8%); the detection mAP of the ultra-small target (the pixel proportion is less than 2%) reaches 65.7%, and the index of the traditional algorithm is generally lower than 40%.

The invention discloses an aerial target detection method based on dynamic attention and double-frequency feature enhancement. The method comprises the following steps: 1) extracting multi-scale features by using a backbone network in combination with a low-light feature fusion module, and enhancing detail expression in a low-light and low-contrast scene; 2) sparse modeling is carried out on deep features through a dynamic sparse attention module, key attention connection is reserved, and the global semantic ability is improved while the calculation amount is reduced; 3) inputting the deep enhanced features and the shallow features into a double-frequency feature enhancement module, and respectively modeling low-frequency background and high-frequency details to realize foreground highlighting and background suppression; according to the unmanned aerial vehicle aerial image target detection method, the problems of low illumination, complex background and small target detection are effectively solved, and the precision and robustness of unmanned aerial vehicle aerial image target detection are remarkably improved.

The invention provides an unmanned aerial vehicle remote sensing monitoring and evaluating method and system for field orchard diseases and insect pests, and relates to the technical field of orchard disease and insect pest monitoring and evaluating. The interference of complex background noise is effectively eliminated by constructing a ternary separation index integrating feature enhancement, background suppression and brightness normalization; a threshold value is determined based on the exponential distribution, pixels are accurately divided into a non-canopy background, a shadow canopy and an illumination canopy, and fine segmentation of an unstructured scene is achieved; constructing a gain compensation item by using a local illumination reference mean value, correcting a shadow canopy reflectivity curve, and eliminating illumination non-uniform interference; selecting a local high quantile to establish a health reference basis, coupling a discriminant item and an attenuation item, and resolving to obtain a physiological stress index, so as to realize acute capture of early weak disease characteristics; and performing power weighted accumulation on the abnormal pixel difference value, outputting a regional hazard level index, and objectively quantifying the overall disaster risk of the region through nonlinear aggregation.

The invention provides an infrared small target detection method fusing multi-scale features and background suppression, and belongs to the technical field of infrared imaging. An improved encoder-decoder network structure is adopted, and the structure comprises three main parts, namely an encoder, a connecting assembly and a decoder. The encoder is responsible for extracting multi-level feature representation from an input infrared image, the connecting component is responsible for transmitting and fusing features among different levels, and the decoder is responsible for combining high-level semantic information with bottom-level detail information to finally generate a target segmentation result. According to the method, the detection precision is remarkably improved, and the omission ratio is effectively reduced; the background suppression capability is effectively enhanced, and the false detection rate is remarkably reduced; better multi-scale feature fusion is realized, and the characterization capability of the target is further improved; the robustness of the detection system is improved on the whole, and the method can adapt to complex and changeable practical application scenes.

The invention relates to an anomaly detection technology, and discloses an anomaly detection method and system for a flexible circuit board patch, and the method comprises the steps: obtaining a patch image in a preset flexible circuit board, carrying out the illumination normalization processing and background suppression processing of the patch image, and obtaining a background suppression patch image, and carrying out abnormal bending identification on the background suppression patch image based on multi-scale gradient fusion and local structure consistency analysis, judging whether the patch in the flexible circuit board has abnormal bending according to a bending identification result, if yes, judging that an abnormal detection result is abnormal bending, and if not, judging that the patch in the flexible circuit board has abnormal bending. And otherwise, calculating an offset coefficient of the normalized patch image based on double-flow alignment and differential significance enhancement, judging whether the offset coefficient is greater than a preset coefficient threshold, if so, judging that the abnormal detection result is abnormal offset, and if not, judging that the abnormal detection result is abnormal. According to the invention, the communication circuit board patch abnormity detection precision can be improved.

The invention belongs to laser stripe image data processing, and particularly relates to a multi-scale gradient gravity center laser center line extraction method and system, and the method comprises the steps: S1, carrying out the feature extraction and enhancement of a laser stripe image, intensifying detail branches through a space channel weight map, obtaining a bounding box positioning result based on depth separable convolution and residual connection, and obtaining a bounding box positioning result; s2, acquiring a prediction center for each column by adopting a prediction method, and searching a gray peak value in a normal adaptive window to realize coarse positioning; the method comprises the following steps: S1, building a local coordinate system and an elliptical window along the coarse center line, and constructing a multi-scale Gaussian pyramid to carry out background suppression and image enhancement, S4, generating a weighted graph through cross-scale fusion, calculating a weighted gray gravity center in each window, and carrying out image enhancement on the weighted graph. And after smoothing, a high-precision laser stripe center line is output. And the accuracy and robustness of extracting the laser center line are improved.

The invention discloses an unmanned aerial vehicle inspection image defect recognition method and system based on machine learning, and belongs to the technical field of image recognition, and the method specifically comprises the steps: firstly carrying out the illumination equalization, background suppression and visual angle correction preprocessing of an inspection image, and then employing a multi-scale deep network to extract shallow texture and deep semantic features, constructing a multi-dimensional defect feature set; introducing a channel attention mechanism into the multi-dimensional defect feature set, focusing a feature vector corresponding to a defect region, and suppressing a feature vector corresponding to background noise to obtain an enhanced defect core feature; inputting the enhanced features into a hierarchical classification network to obtain a preliminary candidate region; extracting morphological characteristics such as area and perimeter of the candidate region, checking by referring to a defect form template, removing false items and correcting the boundary; and finally, integrating the checked and corrected defect position information and the type result and outputting.

The embodiment of the invention provides a near-field MIMO radar data processing method based on multistage background suppression and target enhancement, and is applied to the technical field of radar signal processing. The method comprises the following steps: acquiring a real-time background template and a real-time image frame of a target scene; calculating to obtain differential image data corresponding to the real-time image frame based on the real-time background template and the real-time image frame; performing Gaussian filtering processing on the difference image data to obtain Gaussian filtering image data; performing morphological processing on the Gaussian filtering image data to obtain morphological image data; performing target enhancement processing on the morphological image data to obtain target enhanced image data corresponding to the real-time image frame; and outputting the target enhanced image data. The technical effect of improving the radar imaging precision is achieved.

The application discloses a textile defect detection method and system based on multi-light source dynamic fusion and a double-branch network, and belongs to the technical field of machine vision and industrial detection. The method acquires multi-view images of a textile under coaxial, low-angle and backlight conditions through an encoder trigger line array camera; a local variance calculation and image fusion are performed by using a Softmax dynamic weight algorithm with a temperature coefficient introduced to generate an enhanced background suppression image; subsequently, a first branch lightweight network containing a depth separable convolution and a CBAM mechanism is used to quickly locate a candidate defect area, and then a second branch network is used to finely classify the cropped candidate area, and a Focal Loss is used to solve the long-tail distribution problem. The application effectively solves the problems of easy loss of small defect features in complex texture textiles and large consumption of video memory, meets the demand of real-time online detection of high-resolution images in industrial fields, and greatly reduces the missed detection rate.

This application discloses a method and system for estimating the direction of arrival (DOA) of weak targets based on Riemannian manifold background suppression and adaptive sparse Bayesian learning. The method includes: acquiring time-series signals using a sensor array to construct a series of sample covariance matrices, mapping them to a point sequence on a Hermitian positive definite matrix manifold space; iteratively calculating the background interference covariance matrix using the non-Euclidean geometric properties and logarithmic shielding effect of the Riemannian metric; mapping the background interference covariance matrix back to Euclidean space, adaptively performing background subtraction based on an energy decision mechanism to reconstruct a positive definite covariance matrix to be measured; inputting the covariance matrix to be measured into a sparse Bayesian learning framework, first iteratively recovering the signal power through adaptive mesh refinement sparse Bayesian learning, then performing a closed-loop iteration of subspace noise cleaning while keeping the mesh fixed to recover the sparse spatial spectrum of the target signal; and finally, using local analytical interpolation techniques to eliminate mesh quantization errors and calculate the precise DOA of the target.

The application provides an infrared defect segmentation method and system based on difference prior and double quantile geodesic reconstruction. The method comprises the following steps: collecting a sequence of infrared temperature images of an object to be measured to obtain a temperature three-dimensional matrix; extracting a previous time sequence frame in the temperature three-dimensional matrix to construct a statistical background model, and obtaining a background-suppressed difference sequence image; performing principal component analysis and independent component analysis on the difference sequence image respectively to construct a defect prior map; extracting an effective area of the defect prior map to obtain a stripe-removed image; obtaining a defect saliency response map based on the stripe-removed image; establishing an effective area mask based on the defect saliency response map, setting a double quantile threshold condition to extract a seed area and an allowed expansion area, performing morphological geodesic reconstruction, and outputting a defect segmentation mask after area filtering. The defect segmentation method does not need supervised training and can effectively resist non-uniform heat flow and texture stripe background interference.

The application discloses a double-gas-chamber neutron detector, belonging to the technical field of nuclear radiation detection and particle detection, comprising a gadolinium film neutron conversion layer, two time projection chamber (TPC) modules and an optional energy quantifier module. After capturing thermal neutrons, the gadolinium film generates multi-state particles, and the two TPC modules realize 4π solid angle neutron detection through electronic signal coincidence. The application replaces the scarce 3 He with gadolinium, combines the track reconstruction capability of the double TPC module, significantly improves the neutron detection efficiency and signal-to-noise ratio, and is suitable for nuclear reactor monitoring, industrial material analysis, agricultural moisture detection and deep space exploration and other fields. Through electronic-gamma coincidence measurement and mode recognition, high-precision neutron signal reconstruction and environmental background suppression are realized.

The application discloses a low-altitude radar ground clutter suppression method based on machine learning, generates an RD-A mark sequence representing learning based on an RD-A data cube; generates a physical prior mask set in combination with terrain data, platform attitude information and background Doppler spectrum distribution; obtains a background embedded representation initial value and a target embedded representation initial value; obtains an interactive enhanced background embedded representation and an interactive enhanced target embedded representation, applies an orthogonal constraint term and a mutual information constraint term to the interactive enhanced background embedded representation and the interactive enhanced target embedded representation, and obtains an orthogonal decoupled background embedded representation and an orthogonal decoupled target embedded representation; outputs a background counterfiltering mask for background suppression, constructs a joint filter, and filters the RD-A data cube by using the joint filter to obtain an RD-A data cube after clutter suppression. The application improves small target separation and detection rate.

The invention discloses an unmanned aerial vehicle remote sensing tree species classification method and system fusing a cascade canopy attention mechanism, and the method comprises the steps: obtaining unmanned aerial vehicle RGB image data of a campus scene, carrying out the preprocessing of the data, and obtaining a standardized input image; constructing a classification model fusing the cascaded canopy attention mechanism, wherein the classification model comprises a background suppression module, a fine-grained feature guide module, a channel alignment neck module and a global-local fusion classification head; inputting the standardized input image into a classification model, and obtaining an enhanced canopy image through a background suppression module; inputting the enhanced canopy image into a module carrying fine-grained feature guidance, and extracting fine-grained canopy features; performing channel calibration and feature smoothing on the fine-grained canopy features through a channel alignment neck module to obtain feature mapping of a unified dimension; and fusing the global canopy semantics and the local texture structure in the feature mapping by using a global-local fusion classification head, and outputting a tree species classification result.

This invention discloses a method and device for precise fish feeding based on water surface ripple induction and visual feedback. The method involves artificially generating physical ripples on the water surface to simulate live bait disturbance and induce fish aggregation. Feeding is then initiated, and a real-time video stream is acquired. After image enhancement and feature map construction, a background suppression model is used to remove water ripple interference and accurately extract the salient areas of the feed. Subsequently, the instantaneous feed drift flux crossing the visual warning line and the cumulative loss within the studied time window are calculated. Once the cumulative loss exceeds a preset threshold, it is determined that the fish have stopped feeding, and feeding is stopped. This invention effectively stimulates the fish's appetite for artificial formulated feed through physical means and constructs a closed-loop feedback control mechanism based on the drift flux of uneaten feed. It solves the problems of low fish feeding willingness, high false detection rate, and difficulty in accurately controlling the feeding amount in traditional feeding methods, significantly reducing feed waste and water pollution while ensuring sufficient feeding.

The application discloses a forest musk deer field image few-shot segmentation method based on multi-modal prior and background suppression, first acquires forest musk deer support image pairs and query images, extracts multi-level features through a frozen pre-training visual encoder, and extracts visual markers and forest musk text embedding vectors relying on a CLIP model; a visual corresponding prior map and a text guided positioning prior map are respectively generated, and are fused into a spatial anchor point prior to enhance the saliency of the query target features. Meanwhile, a semantic conflict perception correction module SCRM is designed, an output rectification weight suppresses field background interference, and a purified support feature is obtained. The enhanced query feature and the purified support feature are input into a multi-layer interaction decoder to calculate attention weights, and a hierarchical self-distillation strategy is adopted, the deep attention of the decoder is taken as a teacher signal to guide shallow learning, and model training is completed in combination with a cross entropy loss. The application solves the problem that the forest musk deer contour can be accurately segmented under the condition of few samples, and provides technical support for intelligent monitoring of wild forest musk deer populations.

The present application relates to the technical field of gas detection, in particular to a real-time mass spectrum analysis system for detecting volatile gas components of grain, which comprises a sample spectrum acquisition unit for obtaining original sample spectra at different depths, then weighting each depth original spectrum according to preset weight, and outputting characteristic fingerprint spectrum; a differential background suppression unit for acquiring real-time background spectrum, and using a closed-loop adaptive multi-scale background suppression method to remove background ion signals of the original sample spectrum based on the real-time background spectrum and the characteristic fingerprint spectrum, and generating a net characteristic spectrum without background; and a target VOC determination unit for determining target VOC in the grain pile based on the net characteristic spectrum through a triple verification mechanism. The system improves the recognition sensitivity and quantitative accuracy of the system for low-concentration target volatile compounds, enhances the adaptability and robustness of the system in a complex grain storage environment, and effectively avoids the misjudgment and missed detection problems existing in traditional methods.

The invention relates to the technical field of underwater acoustic imaging and image enhancement processing, in particular to a high dynamic range sonar image enhancement method based on regional adaptive mapping fusion glow, which comprises the following steps of: S1, preprocessing and normalizing an image; s2, carrying out adaptive region segmentation; step S3, carrying out differential brightness remapping; s4, local reinforcement of a shadow adjacent target is carried out; and S5, visual glow generation and fusion. The method has the technical effects that the target-shadow contrast is greatly enhanced; highlighted target detail fidelity and reinforcement are carried out; unification of background suppression and weak signal enhancement is realized; specificity and high efficiency are realized; and a glow fusion mechanism with both information protection and visual guidance is provided.

The invention relates to the technical field of water conservancy project safety detection and termite prevention and control, in particular to a water conservancy project termite exposure feature recognition method and system, and the method comprises the following steps: collecting an RGB image and a multispectral image of a to-be-detected region, and carrying out the multi-modal alignment and background suppression processing, and obtaining an RGB image after background suppression; a small target detection network is adopted to carry out grading detection on the RGB image after background suppression, and a small target bounding box is obtained; carrying out segmentation processing on the verified and screened small target bounding box by adopting a semantic segmentation network to obtain a small target segmentation map; verifying the small target segmentation image by adopting a time sequence association method; and according to the characteristic type of the termite exposure characteristic, the hazard area and the hazard position number, performing hazard grade determination, and generating a detection report containing the segmentation result and the hazard grade.

The application discloses a defect detection method for rail fasteners based on multi-scale feature enhancement, and is implemented according to the following steps: step 1, selection and preprocessing of an existing rail fastener defect dataset; step 2, construction of a feature extraction module of a rail fastener defect detection network; step 3, construction of a multi-scale feature enhancement and fusion module for rail fastener defect detection; and step 4, construction of a detection prediction and result output module. Through the function processing procedures of layered feature response highlighting, local texture sensitive enhancement, global context semantic aggregation, inter-layer heterogeneous feature collaborative fusion and dual-domain saliency modulation, the application differentiates and collaboratively enhances the shallow texture information, the middle structure information and the high-level semantic information, thereby effectively improving the representation ability, the category discrimination ability and the background suppression ability of the disease area, and improving the stability and reliability of the detection result.

The application discloses a forest and grassland fire smoke and fire point simultaneous detection method. The method obtains visible light and infrared band images of a target area, registers and synthesizes into multi-band synthetic data; the data is input into a pre-trained multi-modal smoke and fire simultaneous detection network for feature extraction, the network adopts a feature fidelity down-sampling module containing a fixed kernel low-pass filter and average pooling during down-sampling, and is combined with a cross-layer connected background suppression and smoke and fire focusing module, a multi-scale smoke and fire shape perception module and a smoke and fire space information rearrangement amplification module for joint processing; finally, according to the feature channel set, the smoke and the fire point area are respectively located, and a fire monitoring distribution map is generated by superposition. The application realizes the simultaneous detection and mutual verification of multi-modal features of smoke and fire, effectively overcomes the loss problem of small fire points and smoke plume details, and improves the accuracy of early fire warning.

The invention discloses a lightweight soybean pod counting method and system based on structure distillation and foreground guidance, and relates to the technical field of computer vision and agricultural intelligent perception. The method comprises the following steps: constructing a soybean pod density estimation model, wherein the soybean pod density estimation model adopts a main-auxiliary double-branch neural network architecture; designing a structured density supervision quaternary loss function which comprises main branch density loss, auxiliary branch density loss, main and auxiliary consistency distillation loss and background suppression loss; performing training optimization on the soybean pod density estimation model, and based on a structure distillation mechanism and a foreground guidance strategy, enabling the main branch neural network to learn the structure expression of the auxiliary branch neural network in a training stage; and estimating the total number of soybean pods in the to-be-counted soybean plant image by using the optimized main branch neural network. According to the invention, through a double-branch collaborative modeling and structural distillation mechanism, the focusing capability and counting precision of the model to the target area are effectively improved, and soybean pod counting is realized.

This invention proposes an image fusion method and system based on physical priors and a continuous depth field, aiming to solve the problems of mosaic artifacts and insufficient clarity caused by discrete layer selection in existing pathological microscopic Z-stack image fusion. The key technical points are: after sequentially performing adaptive background suppression, frequency domain focusing feature extraction, and edge-preserving energy smoothing on the Z-stack image sequence, a continuous depth field is obtained through sub-pixel level interpolation fitting; finally, a high-resolution pathological image with full depth is output based on weighted fusion of the continuous depth field. This invention can be applied to biomedical pathological imaging scenarios, effectively improving the quality of microscopic image fusion, eliminating inter-slice artifacts, and providing high-fidelity full depth image data for pathological analysis.

The invention relates to the technical field of nondestructive testing, in particular to a nondestructive testing method and system for the performance of a vacuum heat insulation composite heat preservation material.The nondestructive testing method comprises the steps that the surface of a vacuum heat insulation composite heat preservation material to be detected is heated, and time sequence infrared image data are collected; performing statistical analysis on an image sequence before heating in the time sequence infrared image data by adopting a dynamic multi-baseline background suppression method to obtain a temperature difference sequence after background suppression; performing feature extraction on the temperature difference sequence after background suppression to obtain a heating rate feature image, a cooling rate feature image, a lifting asymmetry feature image and a time domain feature image; fusing the images to obtain an enhanced defect feature image; and carrying out segmentation processing on the enhanced defect feature image, and outputting a nondestructive testing report. According to the invention, high detection sensitivity can be maintained under different temperature difference conditions, and reliable identification of defects under a low signal-to-noise ratio condition is realized.

The invention belongs to the technical field of image segmentation, and discloses an edge prior Mama-based image segmentation detection method, electronic equipment and a storage medium, and the method comprises the steps: preprocessing an input feature through a space interaction enhancement (SIE) module; sS3D is utilized to capture spatial dependence and depth dimension context in the slice, and long-range dependence is calculated through a state space model; meanwhile, an EP component is used for explicitly enhancing local edge and texture features through convolution and background suppression operation; and finally fusing the outputs of the SS3D and the EP. According to the method, the advantage of Mamba efficient modeling long-range dependence is reserved, meanwhile, the detail perception capability of boundaries is remarkably improved, reliable separation of abnormal areas and precise description of local details under complex background interference are achieved, and segmentation robustness and precision in tasks such as industrial defect detection and medical image analysis are effectively improved.

The invention discloses a visual detection method and device, computer equipment and a storage medium. The image quality is effectively improved through the preprocessing steps of illumination compensation, denoising, background suppression and the like. The local gray scale difference of the potential target area is further amplified through contrast enhancement; a pre-trained convolutional neural network is introduced, the convolutional neural network adopts a YOLOv5 network structure, and multi-scale feature extraction, residual enhancement and feature fusion are combined, so that the system can capture target visual features of different sizes and different depth levels at the same time. The extracted multi-level effective feature vectors are used for training a classification model, and the system can gradually form the target recognition capability for a specific machining workpiece. In a real detection stage, the model quickly deduces the to-be-detected image, and can maintain high detection precision and stability in a complex production environment, thereby meeting the requirements of efficient, real-time and high-reliability industrial visual detection.