233 results about "Computer vision and image processing" patented technology

The invention belongs to the technical field of computer vision and image processing, and discloses a remote sensing small sample target detection method based on double-attention guided transfer learning, and the method comprises the steps: obtaining a remote sensing image data set, and carrying out the preprocessing; taking the preprocessed remote sensing image training set as input, constructing a basic detection model by using a ResNet-101 backbone network, a feature pyramid network and a content awareness upsampling and regional proposal network, and obtaining basic model parameters; basic model parameters are used as input, a DA-FSDET network is trained based on a content awareness strip pyramid and a deformable attention area proposal network, and the trained DA-FSDET network is used to acquire a category detection frame containing small sample categories and confidence. Through cascading and cooperative work of the content awareness stripe pyramid and the deformable attention area proposal network, the detection precision and robustness of the multi-scale target in the remote sensing image are effectively improved.

The invention discloses an RGBT target tracking network and method fused with a multi-interaction feature enhancement mechanism, relates to the technical field of computer vision and image processing, and aims to solve the problems that multi-modal fusion is insufficient and tracking is easy to drift due to fixed or blind updating of a template. The network adopts an end-to-end tracking framework, a backbone network of the network extracts visible light and infrared template images and searches feature tokens of the images through a convolution token embedding module, and performs intra-modal and inter-modal mixed attention interaction by using a multi-interaction Transform module to realize multi-level feature fusion. A target frame is output by adopting an angular point prediction head, a template updating module is introduced, and a template token is dynamically evaluated and updated through two Transform modules, so that the long-term tracking stability is improved. The network effectively deals with complex environment changes by fusing multi-modal information and a dynamic updating mechanism, and the tracking accuracy and robustness are remarkably improved.

The invention discloses a construction scene dynamic obstacle avoidance method and system based on multi-source image fusion, and belongs to the technical field of computer vision and image processing, and the method comprises the steps: mapping multi-source image data to a symmetric positive definite matrix manifold space, carrying out the high-precision registration based on Riemannian geometric measurement, achieving the self-adaptive feature fusion through geometric flow optimization, and achieving the dynamic obstacle avoidance of a construction scene. According to the method, spatial topological features of obstacles are extracted through topological data analysis, and probability trajectory prediction is carried out through a variational inference method. Compared with the prior art, the method has the advantages that the obstacle avoidance success rate is increased by 35%-50%, the false alarm rate is reduced by 40%-60%, the similarity of the technical scheme is lower than 20%, and the method has the advantages that the method is suitable for large-scale popularization and application. And the accuracy, the reliability and the self-adaptive capability of dynamic obstacle avoidance in the construction scene are remarkably improved.

The invention relates to the technical field of computer vision and image processing, in particular to a cross-day-and-night boundary target thermo-optic joint detection method, which comprises the following steps of: firstly, acquiring a visible light image and an infrared image and unifying the visible light image and the infrared image to an image plane reference system; extracting topological features and multi-scale energy features from the input, fusing the topological features and the multi-scale energy features to generate a feature map, and outputting a central heat map, target size regression and sub-pixel offset by adopting anchor-frame-free detection; implementing optimal transmission correction according to an evidence field obtained by normalization of a cross reconstruction residual field to obtain a correction heat map and an initial candidate; triggering a fixation area by using the shape correction heat map and the initial candidate, executing super-resolution and secondary detection in the area, and performing affine reprojection and primary detection fusion to form an updated heat map and a fusion candidate; and in combination with uncertainty and topological consistency, a final detection set is output by using a conditional random field and non-maximum suppression. The method is stable in low-contrast, small-target and strong-interference scenes.

The invention relates to the field of computer vision and image processing, and discloses an image target recognition system based on a convolutional neural network and a feature fusion technology. Comprising a quality evaluation and alignment unit, a reversible decoupling and gating recharge unit, a multi-branch feature extraction unit, an evidence fusion unit, a marginal contribution gating unit, a topology consistency and boundary refinement unit, a detection and positioning unit and a linkage control unit. Reversible decoupling of contents and degradation components is realized in a feature domain, and bitwise gating recharge is implemented in a candidate region according to a quality map, so that small target and weak texture features are enhanced. The system fuses multi-branch output based on an evidence theory, determines weight distribution by combining marginal contribution calculation, triggers gating enhancement and local refinement when a conflict or deviation exceeds a threshold value, and realizes cooperative control of feature suppression and structure correction. According to the method, the recognition stability can be kept in complex scenes such as weak light, blurring and shielding, and the target recognition precision and the system interpretability are improved.

The invention discloses a method for automatically correcting radial distortion of a wide-angle lens, and relates to the technical field of computer vision and image processing, and the method comprises the steps: collecting an original image, calculating the distance from a pixel to an imaging principal point, and extracting a multi-scale image and edge features; analyzing the edge direction and radial change and linear extension of the structure, extracting a linear structure candidate region, obtaining a distortion compensation parameter, generating a multi-scale candidate linear parameter set, and calculating a multi-scale linear consistency index for adaptive adjustment; extracting a point set from the candidate straight line parameter set, executing forward and reverse mapping to calculate a dual-space consistency index, and optimizing local parameters and the point set; calculating a minimum linear complexity index based on the line segment curvature, the curvature change rate and the length, and implementing local optimization; and performing weighted fusion on each index to construct a comprehensive objective function, performing step-by-step shrinkage optimization to obtain an optimal compensation parameter, performing global geometric correction on an original image, generating a distortion correction image, and realizing improvement of the structure recognition precision and the geometric correction effect.

The invention relates to the technical field of computer vision and image processing, and particularly discloses an unmanned aerial vehicle identification and detection method under a target part feature missing condition. The method comprises the following steps: (1) making an initial data set by adopting aerial pictures of an unmanned aerial vehicle; (2) performing labeling and data enhancement processing on the initial data set to obtain a training set and a verification set; (3) a YOLOv8 target detection model is improved, a standard convolution module (Conv) of a backbone network (Backbone) is replaced by dynamic deformable convolution (DEConv), a SimAM attention mechanism is introduced behind a last C2f module of the backbone network and in front of a spatial pyramid pooling layer (SPPF), and in a neck network (Neck), the C2f module is replaced by a C2f-SimAM module, and the standard convolution module is replaced by the dynamic deformable convolution; an additional branch for capturing key features is added in an output branch of a detection head (Head). According to the method, the recognition and detection capability of a target with partial feature missing can be remarkably improved.

The invention relates to the technical field of computer vision and image processing, in particular to a forest wetland environment transition zone identification method and system for an unmanned aerial vehicle. The method comprises the steps of obtaining various prior data layers of a detection area and fluctuation scales corresponding to the prior data layers, taking the fluctuation scales as weights, obtaining similarity degrees between area boundaries of different prior data layers, and identifying similar boundary groups based on the similarity degrees; obtaining the environment fusion degree of the similar boundary groups; and based on the environment fusion degree, identifying an environment transition zone and an unmanned aerial vehicle flight focus region. By fusing prior data layers of DEM, hydrology, vegetation and the like, analyzing boundary similarity according to a fluctuation scale and identifying an environment fusion area, an environment transition zone and an unmanned aerial vehicle key monitoring area are further determined, targeted monitoring and identification of a key interface of a complex ecological crisscross zone are realized, the precision and efficiency of complex environment monitoring are effectively improved, and the method is suitable for popularization and application. And meanwhile, terrain and water area flight risks are avoided.

The invention relates to the technical field of computer vision and image processing, in particular to a foreground detection method fusing wavelet transform and a prompt mechanism, which is characterized in that in a dynamic background, high-speed motion and small target scene, F-measure is averagely improved by more than or equal to 0.30, and the detection precision is obviously superior to that of FFNet and ASCNet. The wavelet transform attention module suppresses background noise, highlights target edges and textures, and improves input feature purity. The CFF, MFR and PB of the decoder collaboratively fuse multi-scale features, redundant information is reduced, and target details and contour integrity are reserved. According to integral end-to-end training, single-frame reasoning only needs 28 ms, and real-time performance and accuracy are both considered. Therefore, the problems that in an existing neural network foreground target detection algorithm, input data are often influenced by noise and background interference, redundant information is easily introduced into a decoder in the multi-level feature fusion and transmission process, and consequently the detection result is inaccurate are solved.

The invention discloses an event camera image reconstruction method of a multi-frame fusion network based on optical flow guidance, and belongs to the technical field of computer vision and image processing. The method comprises the following steps of: 1, acquiring event data, and converting the event data into continuous and smooth space-time voxels by adopting a full-time interval voxel coding scheme based on Gaussian distribution; step 2, constructing an FMF-Net network guided by an optical flow; step 3, designing a comprehensive loss function used for training the FMF-Net network; 4, performing supervised training on the FMF-Net network by using the public event data set; and step 5, inputting space-time voxels by using the trained FMF-Net network, and outputting a high-fidelity reconstructed image. According to the method, long-time motion clues and time domain dynamic features can be fully utilized, high-fidelity image reconstruction is realized under high-speed motion and sparse event input, the problems of loss of reconstruction details and poor consistency of an existing method in a dynamic scene are solved, and the structural similarity and visual quality of a reconstructed image are remarkably improved.

The invention discloses a wall crack detection system and method based on an image recognition technology, relates to the technical field of computer vision and image processing, and generates geometric confidence and constructs a space-time response field by applying micro-vibration excitation and thermal pulse excitation and collecting a visible light dynamic sequence and an infrared thermal response sequence. And the mutual verification gating assembly performs crack candidate region screening based on the opening and closing response and the fusion confidence coefficient of the thermal fracture channel, and forms a crack topology signature evolution sequence through topology signature generation and evolution evaluation. And the closed-loop scheduling component generates uncertainty indexes according to crack evolution characteristics, and dynamically adjusts a detection action parameter set. And updating a fusion confidence coefficient threshold value and a primitive confidence coefficient parameter threshold value through a historical topology difference set and an uncertainty index, and ensuring adaptive updating in a detection process. The system can efficiently and accurately detect wall cracks and provide traceable evolution analysis, and has high robustness and engineering application value.

The invention relates to the technical field of image processing, and discloses a parallelization connected domain analysis method based on neighborhood labeling, computer equipment and a storage medium, through high parallelization design, the steps of father node, root node and area junction point searching, mapping establishing, re-numbering and the like are all designed into an independent parallel computing mode, and the parallel computing mode is designed into a parallel computing mode. According to the method, the acceleration capability of parallel computing hardware such as a GPU can be fully utilized, the processing speed is remarkably improved, the method is easy to expand, large-scale high-resolution images can be efficiently processed, the area of each connected domain is counted by adopting a Hash optimization method of grouping local reduction and global merging, atomic conflicts are effectively reduced, the statistical efficiency is improved, and the calculation efficiency is improved. The robustness of the algorithm in a parallel computing environment is further enhanced, so that the efficiency of the connected domain analysis algorithm is greatly improved, rapid processing of high-resolution images is realized, and powerful support is provided for real-time performance and large-scale application in the field of computer vision and image processing.

The invention discloses an image fusion method based on a double-branch feature decoupling auto-encoder, and belongs to the field of computer vision and image processing. According to the method, for the problems of modal pollution and structural distortion in infrared and visible light image fusion, structural semantic information and high-frequency texture details of a source image are extracted respectively by constructing a content feature coding module and a detail feature coding module, and feature decoupling is achieved. And performing deep fusion on the decoupled features by using an adaptive feature weighting mechanism, and reconstructing a fused image with infrared target saliency and visible light detail definition through a shared decoder. According to the method, an end-to-end two-stage training strategy is adopted for optimization, complex prior or post-processing is not needed, the effects of improving the fused image contrast, edge preservation and target recognition performance are remarkable, and the method is suitable for the fields of weak light monitoring, intelligent perception, unmanned system navigation and the like.

The invention belongs to the field of computer vision and image processing, and provides a weak supervision video anomaly detection method based on vision and text double decisions. According to the method, local and global time sequence modeling modules are constructed, an anomaly focusing visual prompt mechanism and a learnable text prompt vector are introduced, and fine-grained anomaly recognition is realized in combination with a dual-mode memory. In the training stage, the model extracts visual and text features on the premise of no frame-level supervision based on video-level labels, constructs a category alignment graph and optimizes category embedding. In the reasoning stage, the model dynamically updates high-confidence-degree features through a positive and negative memory mechanism, and suppression of prediction deviation between semantic proximity categories is achieved. Compared with the prior art, the method has the advantages that the abnormal behavior recognition capability and the multi-class distinguishing precision of the model are remarkably improved under the weak supervision condition, the structure is simple, deployment is easy, adaptability is high and the like, and the method is suitable for efficient video anomaly detection tasks in intelligent monitoring, behavior recognition and other scenes.

The invention discloses a self-supervised image relighting method based on physical consistency, and belongs to the technical field of computer vision and image processing. The method comprises the following steps: constructing a pairwise training data set containing image and environment map spherical harmonic vectors; constructing an inverse rendering neural network, and decomposing an input image into albedo and geometric visible item spherical harmonic vectors; based on a physical rendering equation, multiple illumination sample pairs in the same scene are utilized to construct a joint objective function comprising reconstruction loss, relighting loss, physical consistency loss and cyclic consistency loss to train the network. According to the method, constant material features are extracted under different illumination conditions through a physical consistency constraint enforcement network, and the robustness of the model under unknown illumination conditions is enhanced through cyclic consistency. According to the method, training can be achieved without the albedo and the geometric truth value of a real scene, the data acquisition cost is effectively reduced, the problem of generalization of synthetic data to the real scene is solved, and high-quality and high-physical-reality-sense image relighting is achieved.

The invention discloses a pedestrian re-identification method based on multi-granularity collaborative prompt learning, and belongs to the field of computer vision and image processing. According to the method, in order to solve the problems that an existing prompt learning method is single in semantic granularity and lacks cross-granularity collaborative modeling, pedestrian appearance details, attribute combination semantics and identity category semantics are respectively represented by constructing fine-granularity prompts, middle-granularity prompts and coarse-granularity prompts. A two-stage training strategy is adopted: in the first stage, a pre-training vision and text encoder is frozen, and loss optimization prompt parameters are monitored through multi-granularity joint supervision; and in the second stage, a prompt and text encoder is fixed, a visual encoder is unfrozen, and the image feature extraction capability is optimized by utilizing pedestrian re-identification task loss. According to the method, multi-granularity semantic information is fused, the discriminant expression ability and cross-scene generalization ability of the model to pedestrian features are enhanced, and the accuracy and robustness of pedestrian re-identification in a complex environment are effectively improved.

The invention relates to the technical field of computer vision and image processing, in particular to a container detection method and system oriented to cross-modal adaptive fusion. The method comprises the following steps: acquiring a visible light image, an infrared thermal imaging image and millimeter wave radar point cloud data in a port scene; constructing a multi-modal adaptive fusion deep learning model containing an environmental perception gating network; the environment sensing gating network analyzes the current environment parameters in real time, and dynamically calculates and outputs the fusion weight of each mode; based on the fusion weight, adaptively fusing the features extracted from each modal; and synchronously identifying the container number and detecting the physical state, the safety state and the operation state of the container by using the fused features. According to the method, the accuracy and robustness of port container identification and state detection under severe weather conditions such as fog, rain, snow and night are remarkably improved, and the safety and efficiency of port automatic operation are guaranteed.

The invention discloses a locomotive detection image filtering processing method and system, and relates to the technical field of computer vision and image process.The locomotive detection image filtering processing method has the advantages that a multi-stage cooperative processing framework is constructed, structural priori knowledge is introduced for intelligent guidance, and compared with simple superposition of a traditional method, various noises are effectively suppressed, and meanwhile, the detection efficiency is improved. Important information such as micro textures and structure edges of key components is protected and enhanced, and the overall definition and detail integrity of the image are remarkably improved; according to the invention, by providing high-quality and low-noise image data, a solid and reliable foundation is provided for a subsequent automatic detection task; according to the method, a closed-loop optimization mechanism of quality evaluation and parameter dynamic feedback is introduced, so that the processing method can automatically adapt to different detection scenes, the processing effect can be self-diagnosed, internal parameters can be dynamically adjusted, and the stability and consistency of output image quality are ensured.

This application relates to the field of computer vision and image processing technology, and discloses a large-model pattern recognition method and system based on deep learning. The method includes: acquiring a temporal image sequence containing a target to be recognized; extracting the geometric topological features of the target in each frame of the image; tracking the target across frames to obtain a temporal numerical set of geometric topological features evolving over time; statistically analyzing the fluctuation parameters of the temporal numerical set within a preset temporal interval, and selecting geometric topological features that meet preset stability conditions from the fluctuation parameters as the essential features of the target; constructing a target pattern representation based on the essential features of the target, and matching the target pattern representation with a pre-stored pattern to determine the recognition result. This application significantly improves the robustness and generalization ability of the recognition model in complex dynamic scenarios such as material changes, illumination changes, pose changes, and even the emergence of new categories, effectively reducing the false positive rate and false negative rate.

An optical remote sensing image processing system and training method, relating to the fields of computer vision and image processing technology, alleviates the difficulties in balancing high accuracy and high efficiency in saliency enhancement in existing optical remote sensing image processing technologies, which suffer from high computational costs, incomplete structures, and unclear boundaries. The optical remote sensing image processing system includes a basic feature extraction network module for extracting basic features from the optical remote sensing image to be enhanced; a multi-directional feature mining and aggregation module for obtaining corresponding feature map sequences based on the basic feature map sequences; a cross-scale edge information fusion module for fusing the feature map sequences through cross-scale edge scanning; and a saliency enhancement module for obtaining a refined saliency map. This invention is applicable to remote sensing and UAV image analysis. It can quickly identify salient areas on the Earth's surface, such as buildings, ships, and disaster areas.

This invention belongs to the field of computer vision and image processing technology, specifically relating to a connector size inspection method and system based on industrial vision. The method includes: determining the effective detection area of ​​the pin through preprocessing and target region segmentation; extracting the original one-dimensional grayscale sequence of the pin's lateral direction and smoothing it to obtain a smooth one-dimensional grayscale sequence; calculating the optical reflection energy integral difference between the two ends of the pin's physical center based on the smooth one-dimensional grayscale sequence to construct a lateral optical reflection asymmetry index; extracting the absolute value of the sum of differences between adjacent rows and the sum of the absolute values ​​of the differences within the evaluation window to construct a structural torsional continuity coefficient; and correlating the maximum value of this index with the structural torsional continuity coefficient to obtain a comprehensive micro-torsion defect index, thereby achieving industrial vision inspection of connector dimensions. This invention effectively solves the problem that traditional two-dimensional inspection techniques struggle to identify pin micro-torsion and have a high false negative rate.

This invention relates to an adaptive severe weather image restoration method based on continuous testing, belonging to the field of computer vision and image processing technology. It includes the following steps: constructing a severe weather dataset and dividing it into training and testing sets; constructing a severe weather image restoration framework, including a pre-trained model DA-CLIP, a student model, and a teacher model; inputting the publicly available dataset into the pre-trained model DA-CLIP for training, and initializing the student and teacher models using the parameters of the trained DA-CLIP model; inputting the training set of the severe weather dataset into the student and teacher models respectively for model training and optimization, obtaining a trained severe weather image restoration framework; inputting the test set images of the severe weather dataset into the trained severe weather image restoration framework to obtain the image restoration result. This invention can improve the model's image restoration performance under complex and realistic weather conditions.

The invention discloses a fine-grained image editing method based on multi-modal thinking chain reasoning, and belongs to the technical field of computer vision and image processing. The invention aims to solve the problem that the existing image editing method cannot meet the requirements of controllability and refined editing at the same time in a complex editing scene. The method comprises the following steps: firstly, generating text chain thinking reasoning according to an editing instruction and an input image by utilizing a multi-modal generation-understanding unified model so as to determine a target object referred by a user; and on this basis, a pixel-level visual positioning image corresponding to the target is generated. Secondly, the model generates semantic reasoning of an editing description and an editing result according to a multi-modal positioning clue, and executes local region editing to generate an accurate edited image; in the training process, positioning enhancement is realized through a multi-modal thinking chain alignment mechanism and auxiliary mask supervision, so that semantic consistency and positioning accuracy between an inference chain and an actual editing area are ensured. According to the method, the editing capability with high interpretability, accurate space alignment and interactivity is realized.

The invention discloses a digital zooming system and method based on deep learning, and belongs to the field of computer vision and image processing. The system is deployed on a microscope and an embedded hardware platform, and carries out dynamic center cutting on a low-resolution microscope image by receiving a zooming instruction so as to reserve a high-information-content area; then, a lightweight super-resolution model is used for processing the cut image, the model integrates a lightweight denoising network, multi-scale feature extraction, improved Pixel Shuffle upsampling and a global-local feature fusion mechanism, high-frequency details are effectively reconstructed, and artifacts are inhibited; and finally, outputting a high-resolution image to a display. According to the method, the multi-scale residual error and the channel attention are combined, key details such as image edges / textures are preferentially reserved, artifacts can be effectively avoided, and compared with a traditional zooming mode, the high-power image is clearer, and more edge details are reserved; according to the method, depth separable convolution and lightweight Transform are adopted, the number of model parameters is small, the reasoning speed is high, and mobile terminal deployment is adapted.

This invention discloses a lightweight image restoration method, relating to the fields of computer vision and image processing technology, comprising: S1, input preprocessing: inputting a damaged image and a size-matched binary mask, concatenating the two along the channel dimension to obtain an input tensor; S2, encoder feature extraction: configuring an encoder network composed of multiple downsampling blocks to extract multi-scale features; the downsampling block includes an LSConv module with improved LSNet convolution, convolutional layers, normalization layers, and activation functions. The LSConv module, with its separable convolutional structure based on multi-branch depth, simultaneously captures global structure and local detail features. Feature correction and fusion are completed through feature concatenation and channel adjustment combined with the SE attention mechanism. This invention innovatively designs a novel LSConv feature extraction module, employing a multi-branch parallel structure combined with multi-scale feature mining, breaking through the limitations of traditional single convolution, simultaneously capturing global and local features, and enhancing feature expression through channel optimization, thereby improving image restoration capabilities.

This invention provides a computer vision-based method for detecting changes in ancient building images, applicable to fields such as computer vision and image processing. The method includes: generating a difference heatmap based on the differences between features of a reference image and features of the image to be detected; extracting multiple geometric cue points based on the pixel values ​​of multiple pixels in the difference heatmap; encoding the geometric coordinates and semantic labels of the geometric cue points respectively, and fusing the encoded coordinates and encoded labels to obtain cue features; using a twin decoding mechanism, simultaneously segmenting the reference image and the image to be detected in the same physical coordinate system based on the cue features, reference image features, and features of the image to be detected, obtaining a reference segmentation mask and its corresponding confidence level, and a test segmentation mask and its corresponding confidence level; and determining the detection result corresponding to the image to be detected based on the reference segmentation mask and its corresponding confidence level, and the test segmentation mask and its corresponding confidence level for each of the multiple geometric cue points.

The invention discloses a hyper-spectral image super-resolution method based on unmixing, which belongs to the computer vision and image processing technology, is used for hyper-spectral image resolution, and comprises the steps of constructing a hyper-spectral image super-resolution framework, carrying out neural network training, and outputting a hyper-spectral image super-resolution framework processing result based on a composite loss function or returning to neural network training. The hyperspectral image super-resolution framework comprises a material component guiding model and a super-resolution model, the low-resolution hyperspectral image is respectively input into the material component guiding model and the super-resolution model, and an enhanced hyperspectral image is output after results of the material component guiding model and the super-resolution model are fused. According to the invention, through a dual-model cooperation framework, the spatial resolution precision of the hyperspectral image and the physical credibility of substance composition analysis are synchronously improved, and joint optimization of spatial enhancement and spectral decoupling is realized.

The invention discloses a low-illumination image enhancement and deblurring combined restoration method in the technical field of computer vision and image processing, and the method comprises the steps: decoupling brightness and color information in an HVI color space, and constructing a double-branch neural network; a dynamic feature perception enhancement module is introduced into an HV branch, and a non-uniform fuzzy region in an image is adaptively processed by generating a position-related dynamic convolution kernel; constructing a structure perception enhancement sub-network, explicitly extracting high-frequency components and edge features of the image, and realizing self-adaptive fusion of multi-source features through a structure guide fusion block so as to reconstruct clear textures and contours; through joint optimization of consistency loss, high-frequency loss and edge loss of HVI and RGB spaces, brightness, color and structure information of an image are recovered cooperatively. According to the method, effective decoupling of the brightness and the color is realized, the non-uniform fuzzy processing capability is improved, the recovery details of the structure and the texture are enhanced, and the robustness is high.

The invention relates to the field of computer vision and image processing, discloses a plant image classification method, and solves the problems that the existing plant image classification technology is poor in generalization ability and low in classification precision, and lacks self-diagnosis classification uncertainty and a continuous optimization mechanism, so that difficult sample classification errors and repetition occur. According to the scheme, the method comprises the steps of obtaining a to-be-classified plant image and performing preprocessing; performing feature extraction on the plant image through a pre-trained feature extraction network to obtain a plurality of image features including different abstract levels; fusing the plurality of image features to obtain a fused feature, and according to the fused feature, generating a classification probability that the plant image belongs to each of the plurality of plant categories; determining an uncertainty metric value of the classification result according to the classification probability; and if the uncertainty metric value satisfies a preset condition, sending the plant image to a labeling terminal for requesting manual labeling of a label, and updating the feature extraction network according to the received manual labeling label and the plant image.

The invention provides a monitoring method and system, a training method, electronic equipment and a computer storage medium, and relates to the technical field of computer vision and image processing. The method comprises the following steps: acquiring an image frame and an event stream of a target scene; preprocessing the image frame and the event stream to obtain a first image feature and a first event feature; aggregating the first event features based on a plurality of preset time windows to generate fused event features; performing cross-modal attention calibration and spatial alignment on the first image feature and the fusion event feature to generate a fusion feature; and reconstructing and outputting an extreme visual image based on the fusion feature and an ambient illumination level estimated according to the image frame and the event stream. The enhanced image with clear details and wider dynamic range is reconstructed under extreme conditions such as extremely dark and overexposure, the perception robustness and target recognition accuracy of the monitoring system under complex illumination are improved, the processing efficiency is considered, and all-weather monitoring can be realized.