32results about How to "Guaranteed classification accuracy" patented technology

The invention discloses a small sample defect classification method based on retrieval enhancement and a multi-modal large model, the method takes similar defect features in different products into consideration, a breakthrough retrieval mechanism is introduced, the mechanism accesses a cross-domain database containing various products and materials, and the retrieval efficiency is improved. In this way, the model learning range is effectively expanded, and the limitation of a single data set or field is exceeded; through the retrieval mechanism, effective training of the classification model can be realized by using small samples; besides, a BNADaster adapter module is designed in the classification model, effective transfer learning is realized through fine tuning of an additional feature adapter, the method is simple and efficient, the calculation cost is saved, and the classification accuracy can be ensured.

The invention belongs to the technical field of artificial intelligence security, and relates to a defense method for model backdoor attacks. The invention aims to solve the problem that the accuracy of the model is remarkably reduced due to the existing defense technology. The method combines a backdoor label detection method and a model fine tuning method, and is realized by the following steps of: firstly, detecting a backdoor label by utilizing a nerve cleaning method, and identifying a potential backdoor trigger; secondly, generating an adversarial sample by using a projection gradient descent algorithm; then, carrying out first-stage adversarial fine tuning training to reduce the sensitivity of the model to a backdoor trigger; and finally, performing second-stage fine tuning training through the clean sample, and gradually recovering the classification performance of the model on the original data.

The invention is applicable to the technical field of knowledge migration, and provides a graph domain adaptive node classification method based on instance learning, which comprises the following steps of: acquiring a source domain data set with a mark and a target domain data set without a mark; the representation of the source domain data set and the representation of the target domain data set are learned through a GNN model, modeling is conducted on the source domain data set and the target domain data set, and embedded vectors of the source domain data set and the target domain data set are obtained; the embedded vectors of the target domain data set are classified through KNN, and target domain category pseudo labels are obtained; performing fine-grained instance contrast learning on embedded vectors of the source domain graph data set and the target domain graph data set; and after classification is finished, displaying a final classification result of the target domain graph data set. According to the method, the defects of an existing graph domain adaptation technology in cross-domain tasks are overcome, and the method has wide applicability and can be applied to graph data analysis tasks in various cross-domain scenes, such as social network analysis, recommendation systems, knowledge graphs and biological networks.

The invention relates to the technical field of image recognition, and discloses a pest identification method and device, equipment and a storage medium, and the method comprises the steps: obtaining a to-be-recognized image; inputting the to-be-recognized images into a pre-trained target identification and recognition model, judging whether the to-be-recognized images are pest images or not, recognizing the to-be-recognized images of which the confidence is lower than a threshold value as other classes, and classifying the to-be-recognized images of which the confidence is higher than the threshold value into corresponding classes; the training process of the pre-trained target identification and recognition model comprises the steps of taking the harmful insect classification data set as a closed set data set, and performing closed set training on the initial classifier model to obtain an initial identification and recognition model; and simulating an open data set by using a composite image generated by the generative adversarial network and a composite feature generated by the feature replicator adversarial training network, and performing open set training on the initial identification and recognition model to obtain a target identification and recognition model. The method is beneficial for improving the intelligent and accurate identification and rapid treatment capability of common pests of imported wood.

The invention provides a multi-level encryption traffic classification system based on deep learning. The data collection module is used for providing original traffic data and providing input for subsequent data preprocessing and model training; the data preprocessing module is used for carrying out standardization processing on the original traffic, ensuring the consistency and stability of input data and providing high-quality input data for subsequent model training; a Globe-Local ResNet backbone network architecture is used for carrying out fusion extraction of global and local features for network traffic, and the detection capability of an original ResNet on the network traffic is effectively improved. And the multi-level classification head module is used for classifying the traffic step by step in a multi-level mode and reducing the problem of sharp increase of the number of categories and the problems of increase of the number of categories and instability of the model. According to the method, efficient multi-dimensional encrypted traffic classification is realized, and through reasonable data processing and model design, the complexity improvement and the accuracy reduction caused by sharp increase of the number of categories are avoided while the classification precision is ensured.

The invention relates to a tongue feature analysis method and device, electronic equipment and a storage medium. A trained tongue detection model and a trained tongue segmentation model are fused to obtain a target segmentation model, an image data set is segmented through the target segmentation model to obtain a tongue segmentation image data set, and the tongue segmentation image data set is labeled to obtain a tongue feature classification data set. Training a tongue feature classification model based on the tongue feature classification data set to obtain a trained tongue feature classification model, obtaining to-be-detected target image data, inputting the target image data into a target segmentation model to obtain tongue segmentation image data of the to-be-detected target image data, and obtaining the tongue segmentation image data of the to-be-detected target image data. According to the method, the trained tongue feature classification model is used for performing feature analysis on the tongue segmentation image data to obtain a tongue feature analysis result, so that the training efficiency of the model is improved, and the classification stability and reliability are also remarkably improved.

The invention discloses a federated learning-based backdoor robust image classification method, which comprises the following steps that: 1, a server initializes a global model and issues the global model to a client; 2, the client initializes a local model based on the received global model, and trains a local neural network based on a local data set; 3, the client side calculates local model update and sends a local update vector to the server; 4, the server records the update vector submitted by the client; 5, the server performs model aggregation according to an improved federal aggregation algorithm; 6, the server issues the trained global neural network to each client; and 7, the client performs image classification by using the trained global neural network. According to the method, federal learning is utilized, and collaborative image classification with higher robustness for backdoor attacks is realized.

The invention discloses a robust training method based on supervised comparative learning, and the method comprises the steps: carrying out the transformation, including rotation, cutting, color jitter or random gray scale, of an image, obtaining two enhanced samples, storing the two enhanced samples, carrying out the PGD or AutoAttack attack of the enhanced samples, and improving a comparative learning method; moreover, the enhanced sample, the corresponding adversarial sample and the sample with the same label in the data set are used as positive samples, other samples are used as negative samples to carry out model training, and supervised contrast loss is used as a loss function to obtain a model with robustness. Experimental results show that the method not only maintains the classification accuracy of clean samples, but also significantly improves the robustness of the model to adversarial samples. The achievement provides a new perspective for improving the safety and reliability of the deep learning model in practical application.

The invention discloses an RSVP normal form electroencephalogram electrode offset self-adaptive compensation method for an autonomous motion scene, is suitable for an electroencephalogram signal decoding task in a motion state, and belongs to the crossing field of computers and biomedical engineering. For the problem of artifact interference in a motion state, positive and negative sample pairs obtained by sampling in an anchoring feature space without artifacts and a feature space to be compensated with artifacts are introduced through supervised contrast learning based on a decoupling training strategy, and the compensation capability of a feature extractor is trained by applying InfoNCE loss. Meanwhile, on the basis of small-batch sample data characteristics, a lightweight feature extractor is designed, and the EEG signals are downsampled into spatiotemporal features based on spatiotemporal convolution for subsequent comparative learning. Compared with a complex feature extractor using a self-attention mechanism, a multi-scale feature extraction module and the like, the lightweight design has the advantages that classification indexes such as BA and RECALL are obviously improved, and the calculation cost is greatly reduced.

The present invention discloses a waste collection device for the production of outer sheath overhead insulated pipes, which relates to the technical field of empty insulated pipe production, and includes a classification table, on both sides of which the outer surfaces of the classification table are fixedly installed with connecting steels for connection. After the waste generated by trimming the outer sheath overhead insulated pipes is transported to the inside of the processing box through a baffle conveyor belt, a crushing and paving mechanism plays a role in crushing it, so that it forms small pieces and falls onto the storage table, and then under the action of the cylinder, the storage table drives the waste to fall onto the classification table, and then the image sampling mechanism collects, compares and processes the waste, and then the robot arm cooperates with the vacuum suction rod to put different waste into different classification chutes in the receiving table so that it finally enters different classification slots. The equipment has a high degree of automation, and can crush and classify the production waste of the outer sheath overhead insulated pipe, with high efficiency and effectively avoiding the problem of waste of resources.

The invention provides a hyperspectral image classification method and system based on a hybrid convolutional neural network, and belongs to the technical field of image processing. Comprising the following steps: carrying out band selection on a historical data set, dividing the historical data set in a Patch form, and taking the divided historical data set as a prepared training sample; determining a training sample based on the proportion setting; training the hybrid convolutional neural network, that is, performing shallow feature extraction and fusion on the images in the training samples by using a context interaction network to obtain a first feature map; inputting the first feature map into a 3D-2D hybrid neural network, and performing deep feature extraction; adaptively adjusting model parameters of the hybrid convolutional neural network in feature extraction to determine an optimal weight; inputting a to-be-classified hyperspectral image into the hybrid convolutional neural network under the optimal weight to perform feature extraction again so as to obtain a new feature map; and inputting the new feature map into a softmax-fc classifier to realize the classification of the hyperspectral image. According to the invention, high-efficiency and high-precision classification of the hyperspectral image can be realized.

The invention discloses a scanning archive digital fine classification method based on image recognition, which comprises the following steps of: acquiring a scanning archive image, and performing layout correction and image enhancement to obtain a normalized image; performing character recognition and layout structure understanding on the standardized image to form a character result and a structure result; constructing a corresponding prototype generation space in combination with the category hierarchy system and the service attribute system; performing hierarchical constraint consistency scoring in the prototype space and generating an evidence packet; and determining category levels and service attributes according to the scores and the evidences, and outputting a fine classification result. According to the method, a method based on image recognition and hierarchical constraint reversible generation consistency evaluation is adopted, digital fine classification of the scanning archives is achieved, and the method has the advantages of being high in classification stability, high in result traceability and high in service adaptation capacity.

The invention provides a high-precision excitation inrush current waveform identification and classification method, and belongs to the technical field of electronic information. The technical problem that the complex excitation waveform is difficult to accurately identify by the existing method is solved. According to the technical scheme, the method comprises the following steps: S1, selecting nine feature variables based on preprocessed data; s2, performing a round of feature screening by using U test for the 9 feature variables; s3, performing two rounds of feature screening by using collinearity test and feature importance sorting; s4, on the basis of RandomForest, GradientBoosting and Light-GBM, a soft voting integrated classifier is constructed, and on the basis of the RandomForest, the GradientBoosting and the Light-GBM, the soft voting integrated classifier is constructed; and S5, carrying out excitation waveform classification based on a soft voting ensemble classifier. The method has the beneficial effects that the excitation waveforms can be accurately identified and classified.

The present invention discloses an electromagnetic signal feature intelligent classification method based on lightweight progressively differentiable architecture search, comprising the following steps: step 1, using the pre-processed IQ samples generated by MATLAB as a proxy data set; step 2, constructing a search space and a unit structure to search for normal units and reduced units; step 3, searching for the best network architecture using a progressively differentiable architecture search method based on step 2; step 4, the training process of the final architecture; step 5, migrating the searched structure to the target task RadioML2016.10B; step 6, retraining to obtain the final model. The method disclosed by the present invention considers the classification accuracy and computational complexity of the model at the same time, achieves a balance between performance and computational overhead, and reduces the complexity of the model without almost reducing the modulation classification accuracy.

The utility model discloses a nut grading, extruding and shelling device which comprises a feeding hopper arranged at the top of the device and used for feeding nuts; the roller is located below the feeding hopper, a continuous grading slide way is arranged below the roller, the grading slide way extends in the axial direction of the roller, and the width of a gap in the grading slide way is continuously increased from the starting end to the tail end; the extrusion roller group is arranged below the grading slide way and comprises a plurality of groups of roller pairs, and gaps of each group of roller pairs correspond to different sections of the grading slide way; and the power system comprises a speed reducing motor and a belt assembly which are sequentially connected and drive the roller and the extrusion roller to synchronously rotate. And the nuts slide along the grading slide way under the action of gravity and are graded through the gradually increased gaps. The classified nuts respectively enter the corresponding extrusion roller pairs to be extruded and husked, and finally, the whole set of structure realizes continuous operation from feeding, classification to husking of the nuts under the support of the rack through the discharge port, so that the classification precision is ensured, and the husking efficiency is improved.

The present invention relates to a slag waste treatment method based on image recognition, comprising the following steps: S1: arranging a plurality of high-resolution image acquisition devices above the slag waste conveyor belt, pre-processing the acquired images, splicing the images at different angles, and obtaining a real-time monitoring splicing image; S2: constructing a slag waste classification model based on a machine learning algorithm; S3: identifying different types of slags according to the slag waste classification model and the real-time monitoring splicing image, and performing preliminary sorting; S4: providing an image acquisition device at each classification collection point, obtaining a real-time image of the classification collection point, and performing classification confirmation based on a slag classification confirmation model; S5: performing secondary fine sorting based on the classification confirmation result, and correcting the recognition error through a feedback loop, and adjusting the sorting. The present invention can not only improve the accuracy and efficiency of slag waste identification and sorting, but also enhance the flexibility and adaptability of the sorting system, and ensure stable operation under different environments and conditions.

The present application belongs to the technical field of power transmission line engineering cableway design and construction, and particularly relates to a power transmission line cableway stockyard point mode recognition method and system based on slope compensation and road width dynamic correction, a storage medium and a computing device. The method comprises the following steps: S1, road feature extraction: including extracting road center line feature data, road final width feature data, and road unit direction feature data; S2, road feature enhancement processing; S3, calculating road curvature; S4, mode recognition classification: using a feature space dynamic weighting segmentation algorithm, through an iterative optimization process in a dynamic weighting road feature space, dynamically adjusting road feature weights, scaling the original road feature space, so that the left and right boundary points in the transformed space can be effectively separated by a hyperplane, realizing accurate two-classification of the stockyard points; S5, space topology consistency enhancement. The method can solve the problem of road width measurement distortion caused by steep slope terrain.

The present invention provides a hyperspectral image classification method, system and storage medium, and relates to the technical field of remote sensing image classification. The present invention uses superpixel segmentation and PCA to process hyperspectral images, obtains a reduced-dimensional hyperspectral image as spectral information, and performs filtering processing on the reduced-dimensional hyperspectral image through multiple edge-preserving filters to obtain multiple filtered images as spatial information, and then classifies the hyperspectral image in combination with the spatial information and the spectral information, making full use of the spectral information and spatial information of the hyperspectral image, with high classification accuracy, and the technical solution is different from the traditional global PCA processing of the hyperspectral image, but uses superpixel segmentation to divide the hyperspectral image into multiple uniform areas, and then performs PCA processing separately, avoiding the loss of image edge information and other independent components, and ensuring classification accuracy.

The application discloses a motor bearing fault detection system and method based on robust deep learning, belongs to the technical field of mechanical fault detection and intelligent sensing, and extracts features of original vibration signals with labels based on a supervised learning branch network, takes the original vibration signals as reference samples, takes the reference samples with the same fault category and different fault categories as positive samples and negative samples, introduces a triplet loss to optimize the inter-class separation and intra-class aggregation relationship in an embedding feature space, and generates embedding representation; based on an unsupervised learning branch network, the original vibration signals after artificial feature extraction based on the time domain and the frequency domain are encoded, a triplet loss is introduced for unsupervised embedding learning, and high-level feature embedding representation is generated; the embedding representation output by the two branch networks is fused, a double loss of the triplet loss and the center loss is introduced for training, and the bearing fault detection model after training is used for bearing fault detection.

This invention proposes a hierarchical label text classification method based on the MOE model, comprising the following steps: encoding the input text using a shared encoder to obtain a unified vector representation of the text; inputting the text vector representation into a router module to calculate the probability distribution of the first-level labels, and activating the top-k expert modules corresponding to the first-level labels; the activated expert modules performing classification predictions within their respective second-level label subspaces; and weighted fusion of the prediction results from the expert modules to obtain the final hierarchical label classification result. This invention extracts text semantic features uniformly through a shared encoder, avoiding redundant learning across branches; it allows samples to simultaneously activate multiple candidate experts through a Top-k soft routing mechanism, breaking the rigid single-path limitation of traditional top-down methods; and it combines routing weights with expert outputs, using the maximization of joint probability as the decision criterion, significantly reducing the number of model parameters and the risk of error cascading while ensuring classification accuracy.

The invention relates to the field of attack and defense confrontation of computer vision, in particular to a method for constructing a confrontation sample defense model based on image preprocessing, and the confrontation sample defense model comprises an auto-encoder for reconstructing a sample and an image classification model for classifying the reconstructed sample; the self-encoder adopts the clean samples to carry out autoregression training, and learns prior distribution of the clean samples; training of an image classification model: firstly, optimizing adversarial samples through an auto-encoder, then training the adversarial samples obtained through optimization, and enabling the adversarial samples to keep distribution closer to that of clean samples in manifold by incorporating distribution information of the clean samples into a training process; the problem caused by neglecting data distribution in the prior art is solved; in the confrontation and defense process, firstly, a sample is reconstructed through an auto-encoder, so that the attack intensity can be greatly weakened, and the robustness can be greatly improved. The method is suitable for various image classification fields, especially high-precision fields such as aerospace and the like.

The invention relates to an intelligent algorithm and device for motion classification and recognition, electronic equipment and a storage medium. The method comprises the steps of synchronously collecting windowed original motion data of multiple parts of a motion object, calculating feature values in a feature engineering framework according to the original motion data, and screening excellent feature values through a mode of calculating information gain. Two original classifiers XGBoost and LightGBM are mixed, the weights of the two classifiers are adjusted in real time based on probability distribution entropy, and a hybrid classifier with adaptive weight adjustment is constructed. A large amount of original motion data is adopted to train a hybrid classifier, a cross validation result is used as a basis to carry out parameter adjustment and optimization on a new classifier, and a classifier most suitable for a current classification scene is constructed. On the basis of the classifier, an outlier confidence screening module is added, a transition stage in the motion conversion process can be accurately captured, motion classification is made to better fit the actual motion condition, and the classification efficiency is improved while the classification precision is guaranteed.

The embodiment of the invention discloses a data classification model training method and device, electronic equipment and a storage medium. The method comprises the steps of performing enhancement processing on a training sample in training data to obtain a first enhanced sample and a second enhanced sample; respectively carrying out feature extraction on the first enhanced sample and the second enhanced sample to obtain a first feature and a second feature; clustering the training data to obtain a plurality of reference categories and a plurality of clustering centers; comparing the first feature and the second feature with each clustering center to obtain a first probability and a second probability; performing feature calculation according to the first feature and each first probability to obtain a first prototype, and performing feature calculation according to the second feature and each second probability to obtain a second prototype; and based on each first prototype and each second prototype, performing prototype comparison learning on the initial model to obtain a data classification model. The classification accuracy of the data classification model can be improved.

The invention discloses a multi-spectrogram lightweight acoustic scene classification method and device for auditory auxiliary equipment, equipment and a storage medium. The method comprises the following steps: firstly, acquiring a hearing-aid front-end environment audio signal to be processed, converting the hearing-aid front-end environment audio signal into a Mel spectrum feature map, a constant Q transformation spectrum feature map and a first-order difference spectrum feature map, and then stacking the Mel spectrum feature map, the constant Q transformation spectrum feature map and the first-order difference spectrum feature map to form a multi-channel spectrum feature map containing complementary time-frequency information as model input; then differential shallow layer feature extraction is carried out through an asymmetric multi-stream module, and depth features are extracted on a low-power-consumption computing platform by using a lightweight backbone network with a ghost bottleneck module and coordinate attention as cores; a spectral morphology module, a frequency band attention module and a channel weight balance fusion module are introduced, and multi-dimensional enhancement and dynamic weighting are carried out on the depth features; and finally, outputting an audio scene classification result by the classifier. According to the method, the classification performance of complex acoustic scenes can be improved while the model parameter quantity and the operation power consumption can be remarkably reduced.

The application discloses a kind of based on spectral gradient field guide's bayesian adaptive graph convolution hyperspectral remote sensing image classification method, utilize the spectral information of pixel in hyperspectral remote sensing image and spatial neighborhood relationship to construct initial feature representation;Introduce spectral gradient field to describe local spectral variation amplitude and direction, adaptively adjust the connection relationship between pixel nodes;According to local spectral difference, dynamically expand or shrink neighborhood range, optimize the expression of graph structure in complex heterogeneous region;With Beta distribution, the connection probability between nodes is modeled, the robustness and stability of graph topology structure are improved;For large-scale hyperspectral scene, construct block reasoning mechanism, divide the image into several spatial sub-blocks and complete graph construction and classification reasoning independently, reduce the computational complexity and storage overhead;Pixel-level classification prediction is carried out on the whole image, and high-precision classification result map is generated.The application enhances the expression ability of strong spatial heterogeneity region, and improves the hyperspectral image classification precision and robustness.

The present disclosure provides a neural network model training method for image processing, which can be applied to the field of artificial intelligence technology. The method includes: performing data enhancement on the source image to be processed, adding the same transparent color and background images of different styles to obtain an enhanced data set; using an edge detection algorithm to extract the edge position of the captured object of each image in the enhanced data set; using a morphological processing method to calculate the boundary area information corresponding to the edge position, and intercepting the captured object information of each image in the enhanced data set according to the boundary area information to form a training data set; inputting the training data set into a convolutional neural network model for training to obtain an image classification model. The present disclosure also provides a neural network model training device, equipment, storage medium and program product for image processing.

The application provides a skeleton behavior recognition video frame extraction method and system under multiple time scales, and the technical points are as follows: first, a target detection algorithm is used to frame the person in the video to obtain the position information of the person in the video; then, a human key point estimation algorithm is used to obtain the key point position of the person information in the video; subsequently, the stacked human key point heat map is uniformly sampled, and the uniformly sampled frame is low, medium and high frequency sampled, so that the model can learn the features under different scales of the video from coarse granularity to fine granularity in a hierarchical manner, thereby enhancing the understanding ability of 3D-CNN for long videos; finally, the convolution channels of the frames under different frequencies in multiple scales are obtained in a parallel manner, the corresponding feature information is obtained, the feature information is subjected to judgment and normalization processing, according to the probability result after recognition, and finally the behavior recognition category is output. The application can improve the performance of the model on long videos while ensuring the accuracy of classification.

The invention relates to the technical field of environmental monitoring, in particular to an urban wetland fine classification method considering different seasonal phase combinations, which comprises the following steps: S1, establishing a wetland fine classification system suitable for urban characteristics; s2, comprehensively collecting wetland sample data, and obtaining a sample point data set; s3, adopting a remote sensing image weighted fusion method to carry out standardized preprocessing on seasonal remote sensing image data of wetland features convenient to extract in the sample point data set, and establishing a multi-seasonal image data set; s4, screening out an optimal feature set according to the multi-season image data set in combination with multi-season features; s5, performing parameter optimization on the pre-training model through the optimal feature set, inputting a multi-season image data set and a preset urban wetland feature library containing typical features of different urban wetlands, and constructing a classification model; and S6, deploying the classification model to carry out urban wetland fine classification work. According to the invention, wetland types can be subdivided rapidly and precisely, and the classification effect is good.

The present invention discloses a hybrid privacy protection image classification method based on federated model distillation, and the steps include: 1. The client pre-trains a local neural network; 2. The central server obtains a public data set and randomly samples it, and then sends it to the client; 3. The client predicts the sent data; 4. The client performs norm clipping on the prediction result and randomly splits it into prediction slices, and sends it to the corresponding client according to the number; 5. The client aggregates the prediction slices to obtain a confused prediction and uploads it to the central server; 6. The central server aggregates the confused prediction and adds noise to obtain a global prediction, and then sends it to the client; 7. The client trains the local neural network by distilling the global prediction, and then retrains the local neural network; 8. The client uses the local neural network to classify images. Based on the federated model distillation algorithm, the present invention uses the secret sharing idea and differential privacy technology to achieve privacy-protected image classification.

The present application provides a method and device for recognizing EEG signals based on a convolutional neural network. An input signal is obtained by preprocessing and data standardization of the EEG signal, and the input signal is input into a conscious movement classification recognition model to obtain a conscious movement recognition result after classification and recognition. The design and construction of the conscious movement classification recognition model can consider the signal components of multiple signal channels in the input signal, identify the conscious movement channel components that reflect the classification of the signal components, and use the channel data of each signal component to merge into an overall one-dimensional data to identify the overall conscious movement component, and then combine the conscious movement channel components and the conscious movement overall component to realize the conscious movement classification recognition of the EEG signal. In this way, the characteristics of each signal component and the overall synergistic characteristics of multiple channels can be considered to improve the accuracy of recognition; and multi-channel recognition can be carried out simultaneously, while ensuring the recognition accuracy, the recognition efficiency can be taken into account.