272 results about "Linear network coding" patented technology

The invention provides an image segmentation method and device combining feature difference recognition and detail enhancement, and the method comprises the following steps: obtaining a to-be-segmented image to construct a data set, and carrying out the marking and preprocessing of the data set; constructing an image segmentation model, wherein the image segmentation model comprises an encoder network, a jump connection network, a decoder network and an output network; the encoder network uses a pre-trained CNN (Convolutional Neural Network) and Transform hybrid network to carry out feature extraction; the jump connection network comprises a channel space interaction module for fusing encoder features and decoder features; the decoder network comprises a convolution-channel attention module for extracting key features and a convolution-up-sampling module for up-sampling; the output network comprises a multi-level feature fusion module for fusing the multi-scale features of the decoder; training the image segmentation model to obtain a trained image segmentation model; and segmenting the image data by adopting the trained image segmentation model.

The invention discloses a denoising network model based on a diffusion model and an oracle rubbing image restoration method, which can improve the universality of the restoration method, improve the diversity of restoration results, reduce the calculation overhead and reduce the manual dependence. The denoising network model is adjusted based on an existing U-Net network and comprises a convolutional trunk network, an encoder, a middle layer, a decoder, a deep convolutional layer and a feedforward neural network which are connected in sequence; segmenting the input image by the convolutional trunk network; the encoder comprises a plurality of groups of cascaded paired visual state space blocks and image merging blocks; the decoder comprises a plurality of groups of cascaded paired visual state space blocks and image expansion blocks; performing down-sampling and up-sampling operations on the image merging block and the image expansion block respectively; the visual state space block adopts the architecture of an existing Mama module, and a two-dimensional selective scanning module is selected to replace an S6 module of the Mama module.

Methods and apparatuses for performing MIMO transmissions are described. Retransmission schemes are described for MIMO transmissions. In an example, a transmitter node obtains a plurality of packets to be transmitted to an intended receiver node over a plurality of MIMO layers. A set of one or more cross-block check blocks is generated, where the set of one or more cross-block check blocks being is using a plurality of bits selected from across two or more packets to be transmitted over respective two or more MIMO layers. The set of one or more cross-block check blocks is transmitted over one or more MIMO layers.

The invention discloses a monocular depth estimation method based on hybrid CNN-Transformer, belongs to the aviation field and the computer vision field, and is applied to the field of environment depth perception in low-altitude flight of an unmanned aerial vehicle. Aiming at the problem that local-global information cannot be effectively extracted by a current depth estimation method, the design content mainly comprises the following steps: improving a visual Transform architecture based on an additive attention mechanism, optimizing feature interaction operation between image blocks, and improving the capability of capturing global semantic information; the method comprises the following steps: designing a deep network encoder mixed with a CNN-Transform module, and effectively extracting local-global feature information; a channel attention mechanism is introduced, and the feature fusion capability of the decoder is enhanced. The method is simple in process, an excellent test result is obtained through training on the flight data set Mid-Air of the low-altitude unmanned aerial vehicle, the extraction capability of local-global feature information is effectively enhanced, and the depth perception accuracy of the monocular depth estimation method in the flight scene of the low-altitude unmanned aerial vehicle is improved.

The invention relates to the technical field of remote sensing image processing, and discloses a remote sensing image adaptive identification method and system for territorial space planning, and the method comprises the steps: obtaining a multi-source remote sensing image data set of a research region, feature extraction, cloud detection, quality evaluation and adaptive preprocessing are carried out; carrying out prototype network coding, calculating a category prototype and probability, and supporting fine tuning of a set; carrying out multi-scale cavity convolution and category scale attention fusion; evaluating the adaptability score of the comprehensive fusion feature map set, and carrying out weighted fusion, classification and normalization; change detection is carried out, stable and change regions are segmented, and time sequence context features are extracted and constrained optimization is carried out; entropy is fused, a boundary is decided, uncertainty is estimated, and weighted fusion is carried out according to a change area; conditional random field optimization, confidence level grading and connected domain identification are carried out; the automation level, the adaptive capacity and the recognition reliability of remote sensing monitoring of territorial space planning are improved.

The invention belongs to the technical field of power systems, and particularly relates to a distribution network load prediction model based on artificial intelligence, which comprises a DM-LSTM neural network model, and a distribution network load prediction method of the DM-LSTM neural network model comprises the following steps: S1: firstly, collecting original data; comprise historical generating capacity, weather data, season information, geographic position parameters and equipment performance indexes; s2, establishing a Deep MLP deep network encoder, inputting the collected original data, and then carrying out initialization and normalization processing on data information; s3, a Sat iotempora l LSTM (Long Short Term Memory) model is built; s4, fusing the output of the Deep MLP deep network encoder and the output of the Sat iotempora l LSTM model to obtain a mixed weight so as to reduce a prediction difference; and S5, constructing a Softmax layer at the tail end of the Sat iotempora l LSTM (Long Short Term Memory) model. According to the method, static and dynamic factors are comprehensively considered, more background information, time sequence modeling capability and dynamic mode recognition capability are provided, prediction errors are effectively reduced, and prediction accuracy and reliability are improved.

The invention discloses a self-supervised learning negative two-way distillation anomaly segmentation method, which comprises the steps of preprocessing a non-anomaly unsupervised anomaly detection training data set, and constructing a simulated anomaly data set; training a student network encoder by combining knowledge distillation and negative learning thoughts according to the simulated abnormal data set; performing multi-scale feature fusion on an output result of the encoder; inputting a feature fusion result into a student network decoder, and training the decoder through a reverse distillation and self-supervision anomaly reconstruction task; performing feature fusion on output results of corresponding layers of the student network and the teacher network by using feature fusion based on a channel attention mechanism; and taking the fused features as input to train a segmentation network so as to accurately predict an abnormal position. According to the method, a bidirectional distillation network structure is constructed, and a self-supervised reconstruction task is combined with a negative learning idea, so that the decision-making ability of the model is effectively enhanced, and the image segmentation accuracy in anomaly detection is remarkably improved.

Provided are a method and an apparatus for operating an NCR in a wireless communication system. The NCR includes NCR-MT and NCR-Fwd. The NCR receives a control message from a base station through the NCR-MT, and performs a forwarding operation through the NCR-Fwd on the basis of the control message. The control message indicates a beam applied to the forwarding operation of the NCR-Fwd and a forwarding resource in a time domain to which the beam is applied. The NCRFwd is in an OFF state in a time resource other than the forwarding resource. That is, if there is no explicit or implicit instruction from the base station, the NCR-Fwd is always expected to be in an OFF state.

The invention relates to the field of network communication, and discloses a multi-source heterogeneous network data cooperative transmission method based on dynamic multi-dimensional evaluation and intelligent disaster recovery, which comprises the following steps: collecting system network parameters of a 5G network and a long-distance wired network in real time through a software definition interface module; performing weighted evaluation on the acquired system network parameters based on a dynamic link selection engine, the evaluation dimensions including real-time bandwidth, transmission delay and current link traffic, and generating an optimal communication link combination scheme; carrying out fragmentation and protocol adaptation on the data by adopting a general packaging framework and an intelligent label technology; distributing multi-path parallel transmission according to a priority strategy; at a receiving end, data recombination and disaster recovery supplementary transmission are realized through network coding and multi-path cooperation; and the evaluation weight is dynamically optimized in combination with reinforcement learning. According to the invention, the problems of rigid link selection, low protocol conversion efficiency and insufficient disaster tolerance in the prior art are solved.

The invention discloses an unmanned aerial vehicle mine resource monitoring system based on multi-modal data fusion, and relates to the field of mine resource monitoring, and the unmanned aerial vehicle mine resource monitoring system comprises a model construction module, an adjustment control module, an atlas construction module, a data fusion module and a monitoring module. A function is introduced for analysis, an unmanned aerial vehicle motion control strategy is obtained; a model is used for entity recognition; a relation triple is obtained through dependency syntactic analysis; coordinate atlas space attributes are fused, a topological relation is constructed, a conflict relation is verified, and implicit relation analysis is completed; the method comprises the following steps: obtaining a mine modular geological map architecture, generating node embedding features by using map attention network coding, associating sensor data with map nodes by using a cross-modal attention mechanism, outputting physical enhancement features, executing monitoring analysis on mine resources, keeping stable unmanned aerial vehicle group control, and ensuring that fused data is accurate.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. In one aspect, one of the systems includes an encoder neural network configured to receive the input sequence and generate encoded representations of the network inputs, the encoder neural network comprising a sequence of one or more encoder subnetworks, each encoder subnetwork configured to receive a respective encoder subnetwork input for each of the input positions and to generate a respective subnetwork output for each of the input positions, and each encoder subnetwork comprising: an encoder self-attention sub-layer that is configured to receive the subnetwork input for each of the input positions and, for each particular input position in the input order: apply an attention mechanism over the encoder subnetwork inputs using one or more queries derived from the encoder subnetwork input at the particular input position.

The invention relates to the field of artificial intelligence, and discloses an E-commerce platform commodity recommendation method based on AI intelligence, and the method comprises the steps: obtaining user environment and behavior data through multi-modal perception, generating a three-dimensional intention vector through space-time convolution network coding, and driving a dynamic strategy selector to configure a recommendation weight; generating an initial recommendation list in combination with the commodity feature library and the user portrait; collecting user interaction feedback in real time, analyzing and outputting a correction vector through a lightweight loop network, dynamically adjusting a strategy, and reordering recommendation results; after a user places an order, a scene complementation generator is started, a complementation commodity sub-list adaptive to a current transaction scene is generated based on knowledge graph multi-hop reasoning, and visual fusion presentation is carried out. According to the technical scheme, accurate intention recognition, millisecond-level strategy correction and dynamic complementary recommendation are achieved, and recommendation real-time performance, adaptability and diversity are improved.

The invention discloses a speech enhancement method and system based on a selective state space model Mama, and the method comprises the steps: carrying out the short-time Fourier transform of a distorted speech, obtaining a time-frequency domain signal, splicing the amplitude, real part and imaginary part of the time-frequency domain signal into a signal of three channels, and enabling the signal to serve as an input signal; an input signal is expanded from three channels to a plurality of channels through a first convolution block of a generator network encoder, different resolution characteristics are extracted through an expanded dense convolution layer of the encoder, a receptive field is increased, and the frequency dimension of the signal is reduced to 1 / 2 of the original frequency dimension by using the last convolution block of the encoder so as to reduce the calculation complexity. The signals expanded by the encoder simulate forward and backward dependency relationships of voice signals under different resolutions through N TF-mamba blocks so as to capture local and global features of long-sequence voice signals, and feature enhancement is realized; and the amplitude and the phase of the enhanced signal are predicted through a complex decoder and a mask decoder to obtain enhanced speech after denoising and reverberation removal.

In a blockchain, transacting a record comprises performing linear network coding on the record or on a data file corresponding to the record, to produce a plurality of coded data parts; and storing at least one of the plurality of coded data parts on the blockchain or storing a proof of knowledge on the blockchain, the proof of knowledge derived from the coded data parts. Linear network coding coefficients might be derived from cryptographic hashes of the plurality of coded data parts. An all-or-nothing transform can use the cryptographic hashes to provide the proof of knowledge.

The invention relates to the technical field of image processing, in particular to a small sample image anomaly detection method based on multi-scale attention and efficient convolution, which comprises the following steps: acquiring small sample image data; the method comprises the following steps: constructing a ResNet network encoder, and constructing decoders of an MSCAM module, an EUCB module and an LGAG module; image features are extracted through an encoder, layer output feature mapping from low levels to high levels is constructed, and different levels of feature information are learned to integrate different levels of semantic information of the image; enhancing feature mapping in the decoder by using a multi-scale convolution attention mechanism; a complex spatial relationship and local attention are integrated by using channel, space and grouping gating attention mechanisms; and output vectors of different levels of the network are spliced. The problem that the accuracy of anomaly detection needs to be further improved due to the fact that features extracted in the early stage of coding are not fully utilized in an existing method is solved.

The invention discloses an image defogging method and device based on deep learning, and belongs to the technical field of image processing. The method comprises the following steps: constructing an image defogging model based on an improved U-Net network, including replacing a down-sampling module in an original U-Net network encoder with a multi-scale enhanced down-sampling module and replacing a jump connection structure in the original U-Net network with a cross attention gating module, and performing iterative training on an image defogging model by using the pre-processed foggy image data set, and defogging the foggy image based on the trained image defogging model to generate a defogged image. By introducing the multi-scale enhanced downsampling module and the cross attention gating module, detail information in a foggy image can be effectively extracted, and the multi-scale feature weight is dynamically adjusted, so that the defogging effect is remarkably improved, and particularly, the defogging efficiency is greatly improved when remote sensing images and common scene images under complex haze conditions are processed. And the method has relatively high robustness and generalization ability.

The invention relates to a knowledge tracking model research method integrating difficulty perception and memory enhancement. According to the method, a graph attention network coding exercise-knowledge point topological relation is constructed, and a difficulty embedding layer and a gating fusion mechanism are combined to realize joint characterization of difficulty features and a topological structure. The difficulty coefficient is innovatively introduced as a dynamic regulation factor of attention weight in memory network updating, the memory intensity of high-difficulty exercises is enhanced, and the discrimination and topological consistency of knowledge representation are improved through a multi-task optimization framework integrating graph structure loss and contrast loss. According to the method, challenges of exercise difficulty perception and knowledge point topological relation modeling are effectively solved, experimental results show that the AUC of the model reaches 0.93 (improved by 7.8% compared with traditional BKT) on an ASSIST2009 data set, the AUC of the model reaches 0.90 (improved by 0.21 compared with KSGAN) on an EdNet data set, objective indexes and teaching scene verification show that the method has remarkable advantages in the aspects of knowledge point correlation modeling and knowledge tracking, and the method is suitable for popularization and application. The method provides an innovative solution for personalized education, and has technical breakthrough and industrial application values.

According to the parameter image generation method and device for shortening the dynamic PET scanning time, the electronic equipment and the storage medium, the 3D image of the time frame of the first 25 minutes of dynamic PET scanning is obtained, feature extraction is conducted on the 3D image, the multi-scale feature map is generated through the network encoder part, the multi-scale feature map is mapped through the network decoder part, and the dynamic PET scanning time is shortened. According to the method, a Ki parameter image for network prediction is generated, the Ki parameter image for network prediction is optimized, the optimized Ki parameter image serves as input of next iteration, the steps are repeated until the set convergence condition is met, and the Ki parameter image meeting the requirement is finally obtained. And the quality of the generated Ki parameter image is beneficial to auxiliary diagnosis, so that the clinical examination efficiency and the comfort of a patient are remarkably improved, motion artifacts which are easy to occur during dynamic PET scanning are reduced as much as possible, and the quality of the generated Ki image and the diagnosis accuracy of a doctor are further ensured.

The invention relates to the technical field of computer vision and target detection, and particularly discloses a lightweight real-time two-wheeled vehicle helmet detection method. According to the method, firstly, a video stream is collected and preprocessed through a traffic monitoring camera, then feature extraction and fusion are carried out through a lightweight backbone network, an encoder and a neck network in sequence, finally, a detection result is output through a decoder, a StripCGLU module is introduced to reduce the calculation complexity, a Pola Former module is adopted to enhance the feature interaction capability, and finally, the detection result is output through a decoder. And a GLBiFPN network is designed to optimize multi-scale feature fusion. According to the method, the calculation complexity and parameter quantity of the model are remarkably reduced, the requirement of edge calculation equipment for efficiency is met while high detection precision is guaranteed, and an effective technical means is provided for safety monitoring in an intelligent traffic system.

The invention discloses a deep neural network extraction method suitable for a rock slice image. The method mainly comprises the following steps: adding a light receptive field module behind first five coding layers of a network; according to the network coding layer obtained in the previous step, adding a coordinate attention module behind each light receptive field module; constructing a network decoding layer according to the network coding layer obtained in the previous step to obtain a deep neural network model suitable for rock slice segmentation; using the training data set, and introducing a binary cross entropy loss function to train the deep neural network model constructed in the previous step; and inputting the rock slice image into the deep neural network model trained in the previous step to obtain a segmented image. The rock slice segmentation effect is good, and the method is an effective rock core image particle extraction method.

The invention discloses a contact network insulator damage detection method and device based on few-sample transfer learning. The method comprises the following steps: inputting a to-be-detected contact network insulator image into a two-stage comparison suggestion coding network; the encoder performs feature encoding and feature extraction on the insulator image, and extracts feature representation of the image; the adaptive feature fusion module obtains an optimal fusion weight through adaptive weight learning; a hierarchical feature extraction module receives output of the adaptive feature fusion module, and extracts local position information of small objects in a shallow network of the encoder, global information of large objects in a deep network and local and global intermediate semantic information in an intermediate network layer; the inter-class and intra-class comparison suggestion coding module receives the output of the hierarchical feature extraction module, calculates the semantic similarity between different classes, and achieves the damage detection of the contact network insulator in a complex environment and a high-speed motion scene. According to the invention, the accuracy and efficiency of contact network insulator damage detection are improved.

The invention discloses a knowledge graph dynamic extension method based on knowledge graph and user behavior fusion. The method is specifically implemented according to the following steps: step 1, data preprocessing; step 2, capturing user behaviors; step 3, coding existing nodes in the map, performing feature initialization on the nodes which are not added into the map, calculating the similarity between a user historical behavior vector and all the nodes, screening out K nodes which accord with user interests, and adding the K nodes into a candidate set; step 4, adding new nodes and edges in the candidate set into the existing atlas, and training and updating parameters of the nodes in the new atlas; and 5, outputting personalized path recommendation conforming to the interest of the user, and generating a final map. According to the method, the user behavior feature vector is generated by using the LSTM network to encode the user historical browsing track and the touch click sequence, the nodes conforming to the user interest are screened out through the vector and added into the knowledge graph, and the space-time navigation path meeting the user interest preference is dynamically generated.

The invention discloses a high-frequency perception semi-supervised electromagnetic shielding light window image segmentation method based on multitask and transformation consistency learning, and the method comprises the steps: 1, constructing an electromagnetic shielding light window OM image data set, carrying out the shooting of an electromagnetic shielding light window, selecting an original image, generating a corresponding same-name label image, and dividing the same-name label image into a training set, a test set, and a verification set; 2, constructing a semi-supervised electromagnetic shielding light window image segmentation network model HAMTC-Net; 3, a network model HAMTC-Net is trained; 4, evaluating the performance of the HAMTC-Net network model by using the verification set and optimizing parameters; and 5, inputting an electromagnetic shielding light window image to be segmented into the trained HAMTC-Net network model, and outputting a segmentation result. According to the method, the global learning ability of a network encoder is enhanced through a multi-task method while the transformation consistency is fully utilized, and the high-frequency characteristics of data are fully utilized to guide training, so that the segmentation precision and generalization ability of the model in a complex electromagnetic shielding light window image are effectively improved.

The invention relates to the technical field of mechanical fault diagnosis, and provides a fault diagnosis and classification method for a bearing of an aluminum alloy impeller die-casting ladling machine, which comprises the following steps: acquiring a vibration signal of the bearing of the ladling machine, carrying out continuous wavelet transform on the vibration signal, generating a time-frequency diagram, and carrying out adaptive grid segmentation on the time-frequency diagram. Grid granularity is adjusted according to the local change rate of the time-frequency graph, multi-scale nodes are generated, edge connection is generated for the multi-scale nodes based on a K-nearest neighbor algorithm, and a multi-scale graph structure is constructed; performing unsupervised feature extraction on the multi-scale image structure, including: performing data enhancement on the multi-scale image structure through edge deletion and feature mask to generate an enhanced view; a graph attention network encoder is used for encoding the enhanced view, graph-level embedding is generated, and graph-level embedding is optimized by comparing a loss function; and based on the optimized graph-level embedding, performing fault classification by using a classifier constructed by a graph attention network and a multi-layer perceptron, and outputting a fault category.

The invention provides a three-dimensional seismic data mixed noise suppression method based on an MSAT-Unet. The method comprises the following specific steps: constructing an MSAT-Unet network comprising a multi-scale expansion convolution residual module, a channel-space attention mechanism and a Transform convolution module; the encoder is improved into multi-scale expansion residual convolution, so that the receptive field is expanded, and the capability of capturing local details and global semantic information is enhanced; a channel-space attention mechanism is integrated behind the decoder, a direction sensitive context is extracted through multi-dimensional adaptive pooling, and details and edge recovery are enhanced; meanwhile, an improved decoder is a Transform convolution module, and the feature reconstruction and complex structure recovery capability is improved; the optimized model carries out training and reasoning on the three-dimensional seismic data, mixed noise can be effectively suppressed, a clear data basis is provided for subsequent interpretation, and the method is high in generalization and robustness and good in performance in the aspect of three-dimensional seismic data denoising.

To provide a multicast content distribution system capable of reducing an amount of packet communication in a multicast content distribution based on network coding.SOLUTION: A relay server 2 of a multicast content distribution system is reported by a requesting node of the number of request packets, a transmission capacity with the requesting node, and a total transmission capacity with nodes from which the requesting node can collect encoded packets, and includes a request packet number calculation unit 22 that calculates, as a new number of request packets, a maximum value among minimum numbers of request packets for each requesting node obtained by multiplying the reported number of request packets by a value obtained by increasing a ratio of the transmission capacity with the requesting node to the total transmission capacity by a predetermined premium rate.SELECTED DRAWING: Figure 5

The invention discloses a blast furnace smelting safety reinforcement learning decision optimization method based on expert strategy constraint. According to the method, an optimization strategy can be learned from an off-line expert track on the premise that operation safety is ensured. Specifically, a conditional generative adversarial mechanism is introduced to realize the alignment of strategy distribution and expert decision, and the exploration ability in a security domain is retained while the expert experience is inherited. And an independent state-action safety evaluation network is designed and a discount factor is introduced, so that the long-term accumulation risk caused by the large hysteresis characteristic of the blast furnace is effectively dealt with. In addition, memory is adopted to enhance network coding historical information, incomplete state observation is supplemented, and information deviation in decision is reduced. According to the method, effective integration of triple guidance mechanisms is realized, knowledge inheritance is realized through distributed alignment, performance improvement is driven through reward optimization, and risk prevention and control are ensured through explicit security constraints.

The invention belongs to the field of cryptography and network security, and provides an unmanned aerial vehicle security cooperative computing method based on a certificateless homomorphic network coding signature, which aims at solving the security problem of data transmission in cooperative computing of an unmanned aerial vehicle group, and mainly comprises the following steps: a KGC initializes an Internet of Things data transmission system by generating private and public parameters; the method comprises the following steps of: running a Setup algorithm to generate a bilinear group, a master key and a hash function by taking a security parameter and a dimension of an augmented vector as input; the unmanned aerial vehicle sends an ID to the KGC for registration, and obtains a part of private keys from the KGC; in order to resist pollution attacks, a user divides data into a plurality of vectors before sending a data packet, and generates signatures for the vectors; and after each unmanned aerial vehicle node receives the data packet transmitted by the channel, performing integrity verification on the transmitted data packet, discarding the damaged data packet, and merging the undamaged data packets. And if the node is an intermediate node, sending a vector and signed data packet to a downstream node.

The invention relates to the technical field of intelligent terminal control, in particular to an intelligent management method and system for an integrated first-aid kit, and the method comprises the steps: setting a step length through a long-short-term memory network, synchronously generating a trend change curve through forward rolling and backward rolling, extracting a trend difference, and marking a periodic fluctuation material type under a sensitivity threshold value; fusion of bidirectional time sequence association and sensitivity screening is realized in trend dynamic capture and change identification, the sensitivity of material shortage prediction is improved, and the sensitivity of material shortage prediction is improved by combining resource weight priority allocation and pointer network coding, setting urgency parameters to cut low-frequency categories and reconstructing a material category sending sequence. And hierarchical scheduling and sequence planning of instruction sending are optimized. Based on remote feedback recording of continuous failure categories and reorganization of inventory synchronous update sending instructions, the actual state of the materials is dynamically incorporated into a cyclic supplement process, the continuous supply guarantee capability of emergency material use is enhanced, and the abnormal response speed and the inventory health management level are improved.

The invention discloses a reverse knowledge distillation-based cross-stage feature fusion method for industrial anomaly detection and positioning, and relates to the technical field of knowledge distillation and transfer learning. The method comprises the following steps: step 1, constructing a teacher network encoder, a student network encoder and a decoder; 2, extracting input image features through a student network encoder, and reconstructing the features through a student decoder; step 3, student network cross-stage feature fusion design; 4, locally sensing a dynamic attention module LDA, and adding a sliding window and convolution to replace original global pooling; and step 5, a self-adaptive multi-scale feature fusion module CAAMS-FF establishes a dependency relationship of a context so as to realize a fine-grained high-quality feature reconstruction effect. And step 6, inputting an anomaly graph obtained by the teacher-student network into the segmentation sub-network to obtain a final anomaly detection and positioning result. While the calculation efficiency is maintained, the spatial positioning precision and the multi-scale adaptability are significantly improved.