1515 results about "Feature Dimension" patented technology

The invention discloses an image super-resolution reconstruction method based on deep learning, and belongs to the technical field of image signal processing. Comprising the following steps: S1, realizing multi-dimensional feature representation of a low-resolution input image by using a deep learning network; s2, realizing accurate alignment and lossless mapping of cross-scale features, and effectively solving the problems of feature dimension inconsistency and information distortion; s3, explicitly learning an increment mapping relation between low-resolution input and a reconstructed high-resolution image, and relieving the gradient disappearance problem of a deep learning network structure; s4, enhancing the naturalness and detail fidelity of the reconstructed image through adversarial training; s5, integrating pixel-level reconstruction loss, perception loss and structural similarity loss, and effectively balancing objective indexes and subjective visual quality of the reconstructed image; and S6, realizing joint optimization of deep learning network parameters through a unified gradient back propagation mechanism, and finally outputting a high-quality super-resolution reconstructed image.

The invention relates to the technical field of data fusion, and discloses a multi-source heterogeneous data fusion method and system based on edge computing, and the method comprises the steps: obtaining a heterogeneous data stream, carrying out the data type recognition and data feature extraction, and obtaining an original feature set; according to the original feature set, unifying feature dimensions and adjusting a time reference to obtain time sequence vector data; according to the time sequence vector data, filling the feature value of the missing time point to obtain a multi-modal feature; performing block storage on the multi-modal features, verifying the synchronism of adjacent modals, distributing modal synchronization weight coefficients, and finally generating a fusion feature matrix; according to the fused feature matrix, identifying and filtering redundant feature dimensions, establishing a feature association map, and executing feature merging to obtain a simplified feature matrix; according to the simplified feature matrix, feature importance scores are calculated, sorting weight coefficients are arranged and distributed in a descending order according to the scores, and a multi-modal fusion semantic vector is generated. The method improves the accuracy of data analysis and decision.

The invention discloses an intelligent partial discharge online monitoring and fault diagnosis system based on multi-sensor fusion, and the system comprises a multi-sensor collection module which is used for synchronously collecting data in a partial discharge process; the data preprocessing module is used for preprocessing the partial discharge signal data; the feature fusion module is used for constructing a fusion weighted feature matrix; the feature dimension reduction module is used for constructing a fusion feature matrix after dimension reduction; the partial discharge classification model training module is used for constructing a partial discharge type classification model by adopting a lightweight capsule network; the hyper-parameter search optimization module is used for optimizing the partial discharge type classification model; the classification model deployment module is used for deploying the optimized partial discharge type classification model; and the online reasoning and fault diagnosis module is used for receiving data in real time, generating a fault alarm signal and recording and returning fault event information. According to the invention, a real-time partial discharge on-line monitoring and intelligent fault diagnosis scheme is provided for equipment.

The invention provides a multi-modal document understanding model, a training method, a reasoning method and equipment, global visual features are extracted by using a weight-frozen first visual encoder, the understanding ability of the model to natural scene images is enhanced, a second visual encoder extracts fine-grained features based on high-resolution document images and region-of-interest information, and the understanding ability of the model to natural scene images is enhanced. And the analysis precision of the complex document is improved. And the information interaction module improves the intelligent understanding ability of a specific area in combination with the position of the region of interest input by the user. The feature fusion module splices multi-modal features in a channel dimension, so that visual information from different sources is efficiently integrated. The linear layer converts feature dimensions, so that the visual features are adaptive to the input requirements of the large language model, and the large language model combines visual and text information to generate a text understanding result conforming to semantic logic. According to the model, the capability of analyzing and extracting the fine granularity of the document information is improved by combining the two-way visual encoder with the selection of the region of interest of the user on the document image of the image-text structure.

The invention discloses a farmland abandoned land identification method based on multi-source remote sensing data and time sequence correction. The method comprises the following steps: 1, data acquisition and preprocessing; 2, extracting a key phenological period; 3, constructing a feature library; 4, training and predicting a random forest model; and 5, post-processing and precision evaluation are carried out. The method comprises the following steps: firstly, integrating multi-source remote sensing image data, and strictly preprocessing the multi-source remote sensing image data to ensure the time-space consistency of the data; and accurately extracting key phenological period information by deeply analyzing the minimum vegetative period and the vegetative peak period of the vegetation through NDVI and NDSI time sequences. In the feature construction stage, multiple spectral indexes are fused, principal component analysis and feature importance screening are applied, feature dimensions are optimized, and model performance is improved. A random forest model is adopted for training and prediction, and a high-precision abandoned land spatial distribution diagram is finally generated by combining an oversampling technology and a category weight adjustment technology.

The invention discloses a space-frequency domain graph learning-based adaptive deep fake face detection method and system, and the method comprises the steps: randomly extracting an image frame from a video, intercepting a face image, adjusting the feature dimension of the face image, and transmitting the face image to a depth adaptive wavelet module and a normalized residual homomorphic composition neural network module; a depth adaptive wavelet module extracts frequency features of the face image; a normalized residual homograph neural network module extracts spatial domain features of the face image; performing weighted fusion on the frequency domain features and the spatial domain features by using a self-adaptive feature fusion module based on gated convolution, realizing class attention guidance by using the gated convolution, dynamically adjusting the channel of a feature map and the weight of a spatial dimension, and finally obtaining fusion features; and performing classification according to the fusion features by using a classifier. According to the method, the extraction capability of forged detail clues is enhanced, and the detection precision and stability of the model are remarkably improved.

The invention discloses a large-scale sample storage management method and system based on a multi-dimensional label system. The method comprises the following steps: acquiring large-scale sample data input, and establishing a multi-dimensional label system for each piece of sample data based on a sample source, sample quality and sample feature dimensions; performing automatic auxiliary identification and classification and quality evaluation on the sample data, generating a multi-dimensional label for the sample according to a multi-dimensional label system, selecting a proper storage layer according to the multi-dimensional label of the sample, and storing the sample in a specified storage layer after the storage space is optimized; and returning a retrieval result according to a retrieval condition input by a user, and distributing the retrieved sample to a required task or analysis scene according to a requirement. According to the method, by establishing the multi-dimensional metadata labels, retrieval, positioning and management of the samples can be completed quickly and efficiently. Meanwhile, automatic auxiliary classification and recognition of the samples are achieved, and the data management efficiency is improved.

The invention relates to the technical field of glass detection, and discloses a method, a device and equipment for detecting glass after light sweeping. The method comprises the following steps: collecting multi-source original sensing data of glass to be detected, carrying out signal denoising, synchronous transformation and multi-scale glass feature extraction on the multi-source original sensing data, and carrying out feature combination, feature dimension reduction compression, region segmentation and region contour reconstruction on the multi-scale glass features to obtain an initial defect region set; performing spatial-temporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set; and constructing a region similarity matrix corresponding to the enhanced defect feature set, performing defect category judgment and multi-dimensional glass quality scoring on the region similarity matrix to obtain glass quality evaluation data, performing process association on the glass quality evaluation data, and generating a final detection report of the to-be-detected glass. According to the method, the result accuracy of quality detection on the glass subjected to light sweeping is improved.

The invention, which relates to the technical field of intelligent interaction, discloses an AI-based adaptive scene intelligent interaction system comprising a multi-modal data acquisition module, a modal preprocessing module, a multi-modal embedded coding module, an intention fusion and representation module, a service scene matching module and a service execution and reinforcement learning module. The method comprises the following steps: acquiring multi-modal original data in a user interaction process, including voice signals, text input and user behavior tracks, and synchronously recording an acquisition timestamp; according to the method, through a multi-modal unified embedding and dynamic weighting mechanism, the problem of characteristic dimension imbalance is effectively solved, and the user intention recognition accuracy is improved; meanwhile, reinforcement learning and a multi-factor scoring model are combined, personalized scene matching and dynamic response are achieved, the adaptive capacity and service accuracy of the system in a complex environment are improved, and therefore the stability and user experience of the intelligent interaction system are remarkably optimized.

The present disclosure relates to the technical field of smart power grids. Disclosed are a method and apparatus for constructing a high-order tensor network of a large-scale power grid, and a device and a medium. The method comprises: acquiring multiple category attribute sets of heterogeneous nodes in a large-scale power grid; performing feature extraction on the multiple category attribute sets, so as to obtain attribute features of the heterogeneous nodes; on the basis of the attribute features of the heterogeneous nodes, using a deep hash mapping model to establish multiple feature sub-spaces; using a breadth learning strategy to align the multiple feature sub-spaces to a unified feature dimension; on the basis of an attribute feature vector of the unified feature dimension, using a distance measurement method to calculate a distance measurement value between the heterogeneous nodes; on the basis of the attribute feature vector of the unified feature dimension and the weight of a heterogeneous node interaction relationship, determining a basis tensor for representing the heterogeneous node interaction relationship; and on the basis of the basis tensor and the weight of the heterogeneous node interaction relationship, using a tensor multiplication operation rule to construct a high-order tensor network of a large-scale power grid.

The invention discloses a multi-scale feature fusion fault prediction method for a high-speed rail bearing. The method comprises the following steps: acquiring a fault vibration signal of the high-speed rail bearing through a vibration signal comparison technology; noise and interference signals of the high-speed rail bearing fault original vibration signals are removed through an adaptive filter; performing multi-level decomposition on the vibration signal by using wavelet packet decomposition, extracting sub-band signals of different frequency bands, calculating instantaneous frequency and energy distribution of the sub-band signals by using Hilbert-Huang transform, and forming a time-frequency-energy three-dimensional feature map; carrying out weight distribution on each feature dimension in the three-dimensional feature map by adopting a self-adaptive weighting algorithm; constructing a training data set through the time-frequency-energy three-dimensional feature map and the fault type label; and training the long short-term memory network LSTM by using the training data set, constructing a fault prediction model, and performing fault prediction on the high-speed rail bearing. According to the invention, accurate prediction of the high-speed rail bearing fault is realized, and the accuracy of fault detection is improved.

The invention relates to a causal effect dynamic estimation method and device and a medium. The method comprises the steps that an intervention perception convolution module is constructed through a longitudinal observation data set, weighting processing is conducted on time-varying covariables and intervention variables, and the intervention variables are embedded to generate multi-dimensional time sequence feature representation. A radial basis function is used for mapping intervention variables to generate high-dimensional feature representation, the high-dimensional feature representation is combined with time sequence features to construct a multi-layer intervention sensing module, and after feature dimension matching is adjusted, time sequence feature representation containing historical features and intervention information is output. And constructing a prediction model on the basis, predicting a result variable and a time-varying covariable of a next time step, minimizing loss by using an SAM optimization algorithm in combination with an observation value, introducing adversarial training to generate a sample, and iteratively optimizing model parameters. And performing anti-fact reasoning by using the model, generating anti-fact trajectories of each time step result variable and time-varying covariable, comparing a plurality of intervention schemes by combining different intervention variable sequences, outputting dynamic causal effect trajectory data, and realizing personalized causal effect estimation.

The invention relates to a multi-sensor fault feature fusion diagnosis method for electrical equipment, which comprises the following steps: synchronously acquiring operation data of the electrical equipment through a vibration sensor, a temperature sensor, a current sensor and an ultrasonic sensor, dynamically adjusting the sampling frequency according to the physical characteristics of each sensor, and the sampling rate of the temperature signal is not lower than 1Hz. Through a multi-source sensor data synchronous acquisition and time sequence alignment technology and a signal alignment method combining a dynamic time warping (DTW) algorithm and Hilbert-Huang transformation, the problem of time asynchronization of heterogeneous sensor data such as vibration and temperature is solved, so that the time alignment precision of multi-source data is improved, the feature extraction accuracy is improved, and the accuracy of feature extraction is improved. Through hierarchical feature extraction and graph convolutional network fusion, a feature incidence matrix based on mutual information is constructed, deep correlation between vibration signal TKEO features and cross-modal features such as current harmonics is mined by using GCN, the feature dimension is reduced, and the fault feature separability index is improved.

The invention discloses a missing perception prompting method and system based on modal specific and general information, and belongs to the field of multi-modal learning. Performing modal missing processing on the initial features of the original video sample; unifying feature dimensions and sizes of modal missing features; wherein after unifying the feature dimensions, splicing to obtain a multi-modal joint feature; performing size unification and then performing fusion to obtain a modal universal prompt; using a gating fusion mechanism to control the fusion of the learnable matrix corresponding to the modal to obtain a modal specific prompt; fusing the modal general prompt and the modal specific prompt to obtain a specific-general prompt as a missing perception prompt; and the specific-general prompt and the multi-modal joint features are spliced to serve as input of a multi-modal model, and a full connection layer is introduced to complete a multi-modal downstream task. According to the method, the learnable specific-universal prompt of mode missing perception is introduced into the multi-mode model, so that cross-mode application in a data mode incomplete scene is realized, and the performance of a multi-mode downstream task is improved.

The invention discloses a cross-modal alignment method and device for a multi-modal large language model, and relates to the technical field of multi-modal alignment, and the method comprises the steps: obtaining an image, carrying out the target detection, and generating a local image fine-grained feature; obtaining a text, extracting text features, extracting a text entity relationship, and obtaining an entity relationship dependency tree as a text entity relationship graph; performing local fine-grained semantic alignment on the local image fine-grained features and the text features; using the scene graph model to obtain a visual relation graph between detection targets in the image; performing graph structure alignment and graph semantic alignment on the text entity relation graph and the visual relation graph; and fusing the semantic alignment features and the graph alignment features to obtain semantic alignment features with rich fine grit, global semantic alignment features and richer relationship alignment features, and inputting the semantic alignment features, the global semantic alignment features and the richer relationship alignment features into a multi-modal large model. According to the method, alignment of different modal data on different feature dimensions is realized, so that the different modal data are accurately associated.

The invention discloses a target detection method based on multi-modal data fusion and 3D voxel projection, and the method comprises the steps: firstly, extracting and fusing multi-modal features from an RGB image and a depth image; thirdly, aggregating the multi-scale features by using a feature pyramid network, and accurately mapping the 2D feature map into a 3D voxel space by using real depth information to form a 3D voxel volume; then, local 3D features are extracted through a voxel feature coding network, and global features are further extracted through a 3D convolutional neural network; and finally, inputting the global features into a detection head to obtain a final target detection result. The model is trained on a large public data set and can adapt to data with high feature dimension, large structure difference and unbalanced distribution, the accuracy and robustness of 3D target detection can be effectively improved through fusion of multi-modal features, and the method is suitable for object detection tasks in a logistics sorting scene.

The invention relates to the technical field of offshore wind field operation and maintenance, and provides multi-modal and multi-source data fused deep learning fault diagnosis, which is characterized in that multi-modal data and multi-source data are acquired, cleaning work, unit conversion and format conversion work are carried out on the data, the data are integrated, and the fault diagnosis efficiency is improved. The time domain features, the frequency domain features, the time frequency features and the text features are taken as basic features to perform data feature extraction work, under the assistance of reducing feature dimension work, the calculation complexity is reduced, useful information is kept, an original model is generated through fusion, the original model is trained and optimized by using a deep learning model, and the calculation efficiency is improved. And fault diagnosis of offshore wind field operation and maintenance is completed for the fault diagnosis model. Multi-modal and multi-source heterogeneous data are combined, various information sources in a wind driven generator system are fully utilized, more accurate and comprehensive diagnosis services are provided, the accuracy and reliability of fault diagnosis are improved, the robustness of the system is enhanced, and therefore stable operation and efficient maintenance of the wind driven generator are guaranteed.

The invention relates to the technical field of plant growth-promoting bacterium screening, in particular to a thunberg fritillary bulb growth-promoting bacterium colony screening method and system based on image recognition. The screening method comprises the following steps: establishing a three-dimensional gradient dilution system of a thunberg fritillary bulb rhizosphere soil sample, and collecting a bacterial colony original image under a specific wavelength combination by adopting a multispectral imaging device. According to the method, by establishing a multispectral collaborative imaging mechanism and a dynamic growth modeling system, the technical problems of feature dimension deficiency and space-time correlation fracture in a traditional method are effectively solved. An improved wavelet transform algorithm is adopted to break through the optical interference limitation of an agar matrix, and high-fidelity extraction of bacterial colony intrinsic textures is realized; according to the constructed three-dimensional feature space fusion model, morphological topological parameters, texture entropy distribution and growth differential features are deeply coupled through nonlinear mapping, and the biological correlation of feature representation is remarkably improved.

The invention discloses a multi-modal sentiment analysis method and system based on multi-dimensional perception, and relates to the technical field of sentiment analysis, and the method comprises the steps: inputting to-be-recognized multi-modal sentiment data into a trained multi-modal sentiment analysis model, the multi-modal sentiment analysis model comprises a feature extraction network, a transmembrane interaction network, a multi-dimensional sensing network and a sentiment prediction network; performing feature extraction on the to-be-recognized multi-modal emotion data through the feature extraction network, and constructing aligned text features, aligned audio features and aligned video features; inputting the aligned text features, the aligned audio features and the aligned video features into a cross-membrane interactive network for feature interaction, and determining interactive fusion features; carrying out feature extraction on the interactive fusion feature along a plurality of feature dimensions in sequence based on a multi-dimensional sensing network to generate a mixed multi-modal feature; and performing emotion scoring on the mixed multi-modal features by using an emotion prediction network, and outputting a target emotion score. Based on the above scheme, improvement of sentiment analysis precision is facilitated.

The invention discloses a cross-modal foreground distillation method for multi-view three-dimensional target detection, and belongs to the technical field of automatic driving perception. According to the method, a laser radar or a multi-modal three-dimensional target detection model is used as a teacher model, a multi-view aerial view BEV perception three-dimensional target detection model is used as a student model, and the model performance is improved through multi-scale feature distillation; a space attention mechanism and a foreground mask module are introduced, and the student model can simulate feature responses of the teacher model in different space positions under multiple scales, focus on a foreground target and suppress background noise; the adaptive module ensures the consistency of the multi-scale features in feature dimensions and spatial resolution, and finally, the difference between the features and regression output is measured through loss functions such as mean square error and KL divergence, so that efficient cross-modal and multi-scale foreground feature distillation is realized. Finally, the multi-view three-dimensional target detection precision and efficiency are improved, and meanwhile, the distillation requirements of multi-scale features are fully considered.

The invention discloses a voiceprint clustering method guided by a speaker recognition pre-training model based on short voices, and the method comprises the following steps: obtaining an original voice signal, and randomly combining a plurality of enhancement strategies to achieve data enhancement; performing voice segmentation on the voice signal after data enhancement based on a uniform segmentation mode; extracting voiceprint features of the segmented voice based on an attention mechanism of global time-frequency domain context modeling; and obtaining a clustering result based on K-means clustering and spectral clustering, matching the clustering result with a real speaker tag, performing reverse transmission based on angle-dependent AAM-Softmax loss, and outputting a voiceprint clustering result. According to the method, the influence of environmental interference on feature extraction can be overcome, feature dimensions which are more effective for identity identification can be screened, a clustering output effect which is superior to that of a mainstream algorithm can be obtained with a relatively low parameter quantity, and the robustness under a noise interference condition is improved.

The invention discloses a hydroelectric generating set operation fault data enhancement method and device, and the method comprises the steps: firstly collecting a set vibration signal, selecting a time-frequency transformation method to convert a one-dimensional vibration signal into a two-dimensional time-frequency image, enhancing the feature dimension of the signal, constructing a diffusion feature migration model, gradually disturbing the data distribution to Gaussian noise through forward diffusion, and carrying out the recognition of the Gaussian noise. The method comprises the following steps of: performing inverse denoising to generate simulation data highly similar to a real fault sample, realizing relevance learning and migration sharing of fault features among different working conditions in combination with an adversarial feature migration architecture, and finally evaluating an enhancement effect by calculating similarity among samples, and inputting enhanced data into a fault diagnosis model to verify precision improvement. Through the combination of time-frequency transformation and a diffusion model, sample scarcity and working condition barriers are broken through, a remarkable effect is shown in the aspects of expanding the fault sample scale and enriching the sample dimension, the similarity of generated data and a real sample is improved, the diagnosis precision is improved, and the model generalization ability is remarkably enhanced.

The invention relates to the technical field of intelligent garbage sorting, and discloses a garbage sorting method based on artificial intelligence. According to the method, heterogeneous data fusion is performed on a multi-source garbage feature data set through a semantic alignment module, feature dimensions are unified, redundant items are eliminated, and a standardized feature set is output. And extracting a spatial position identifier and a physical state identifier of the junk entity, and dynamically dividing time slices by adopting a transfer learning model to generate a periodic junk flow data set. The coordinates of the sorting starting point and the sorting ending point are converted into a three-dimensional space grid to construct a migration track topological graph, flow data and physical state identifiers are combined, sorting paths are clustered and analyzed according to material categories, the material flow intensity in a specific period of a target area is quantified, and a dynamic index matrix is generated. According to the method, space-time correlation analysis of multi-modal junk data is realized, and the adaptability of sorting path planning in a complex environment is improved.

The invention discloses a vehicle-road cloud integrated complex time sequence real-time anomaly detection method and system, and relates to the field of Internet of Vehicles, and the method comprises the steps: S1, constructing an initial anomaly detection model; s2, acquiring a training data set; s3, importing the training data set into the initial anomaly detection model, and carrying out training optimization on the training data set to obtain an optimized anomaly detection model; s4, acquiring data to be predicted; s5, performing anomaly detection on to-be-predicted data by using the optimized anomaly detection model to obtain an anomaly detection result; according to the method, time pooling and a channel attention mechanism are introduced for extracting a complex interaction relationship between a feature dimension and a time dimension; a module based on a channel weight adjustment mechanism is added, the calculation complexity of the model is reduced by dynamically adjusting the channel weight, and the expression ability of the model to key features is improved; and the integrated interactive convolution module is used for capturing a dependency relationship in a multi-dimensional space and supporting simultaneous detection of various types of abnormal events.

The invention discloses a PCB appearance defect real-time detection algorithm based on edge calculation, and belongs to the technical field of computer vision and intelligent manufacturing. According to the algorithm, multi-scale feature extraction is performed on a PCB surface image through a multi-modal feature extraction network, feature dimension compression is realized under the constraint of edge computing resources in combination with a lightweight convolutional neural network, and the computing efficiency and the detection precision are balanced. And dynamically adjusting the weight distribution of the local defect features and the global structure features by adopting an adaptive attention algorithm, and enhancing the significance expression of the tiny defects. By fusing a process feature mapping mechanism of a PCB process parameter database and combining a temperature sensitivity coefficient sensed by real-time environment temperature, the suitability of a detection result and a production condition is improved. And a low-delay defect classification decision is realized by using a lightweight classifier.

The invention discloses a body feeling evaluation method based on electroencephalogram-electromyographic signal fusion and dynamic interaction modeling. The body feeling evaluation method comprises the steps that EEG signals and EMG signals in the lower limb movement process of a subject are synchronously collected; carrying out band-pass filtering, artifact removal and wavelet transform processing on the acquired signals, extracting multi-channel time-frequency features, and forming a preprocessing feature matrix; fusing the time-frequency features of the EEG signal and the EMG signal, constructing a multi-modal feature set, and compressing feature dimensions by adopting a sparse coding method; inputting the compressed feature sequence into a neural network model combining a long short-term memory network and an attention mechanism, and carrying out dynamic interaction modeling; and an interaction index sequence is generated based on model output, and an interaction matrix is constructed through a sliding window and Gaussian kernel smoothing processing, so that visualization of brain-muscle interaction strength and dynamic quantification of a proprioceptive function are realized. The invention further provides a system for implementing the method. The method is high in objectivity, high in feature extraction precision and excellent in dynamic modeling capability.

The invention belongs to the field of environment monitoring and computer vision, and particularly relates to a water body color recognition regression method and system based on space-time causality and manifold learning, and the method mainly comprises the steps: carrying out the detection of a current target water body video sequence, extracting a water body region, carrying out the high-dimensional feature dimension reduction of the water body region, and obtaining a water body color recognition result; and performing feature extraction on the water body region through a space-time causal feature learning model, fusing the flow shape learning features and the space-time causal features to obtain fused features, and outputting a finally predicted water body color value. According to the method, end-to-end assembly line design of preprocessing-segmentation-feature modeling-regression is adopted, manual intervention is not needed from video input to color prediction, and through cascade cooperation of five core modules (video preprocessing, water body segmentation, manifold learning, time sequence causal modeling and color recognition), the real-time performance of the system is improved. Full-link automation from environmental interference suppression, feature extraction to result output is realized, information loss of intermediate links is avoided, and recognition efficiency and robustness are improved.

The invention discloses an underground engineering lining disease detection system based on point cloud, and belongs to the technical field of underground engineering detection. In order to solve the technical problems that an existing underground engineering lining disease detection method is low in detection precision, low in automation degree and the like, underground engineering point cloud data to be detected and corresponding position information are collected, and an improved PointNet + + model is adopted for disease recognition. According to the method, original three-dimensional coordinates of a point cloud are expanded into seven-dimensional point cloud data containing coordinates, normal vectors and reflection intensity, the normal vector standard deviation of points in a neighborhood of each candidate point is calculated to serve as local geometric complexity, a local geometric complexity index is fused into sampling distance measurement, then a multi-scale local neighborhood is constructed by combining sphere query, and therefore the multi-scale local neighborhood is obtained. And extracting geometric features, texture features and deformation features by using a PCA feature dimension reduction technology, carrying out feature fusion based on a normal vector weighting mechanism, finally obtaining disease type classification based on a network model, and calculating the size and position of the disease.

The invention discloses a few-sample classification method research based on feature cache optimization, and the method comprises the steps: (1), employing an improved ResNet-based SwAV model as a backbone, and enabling the SwAV model to be used for generating auxiliary features; (2) taking the image data and the text description as original input, and generating text input with rich downstream language semantics by utilizing GPT-3 to serve as text prompt of a CLIP model; (3) through a feature selection method based on feature similarity and difference, the problems of feature redundancy and inaccurate selection in the feature selection process are solved, feature dimensions with high selection value are identified, and normalization and enhancement zooming processing are performed on the features; and (4) utilizing visual contrast knowledge of SwAV, introducing a learnable cache model, and adaptively mixing prediction results from CLIP and SwAV. According to the method, under the CLIP framework without extra training, the accuracy of few-sample image classification is effectively improved through feature cache optimization and a self-adaptive hybrid strategy.

The invention provides a glutenite logging curve identification method and device based on deep learning, and the method comprises the steps: collecting logging curve data, marking the types, and obtaining an original training data set; performing sample generation based on a quantum state mixed generative adversarial network algorithm to realize data expansion; inputting the expanded data into a feature extraction model, and training the feature extraction model; inputting the data after feature extraction into a feature dimension reduction model to train the feature dimension reduction model; inputting the data subjected to dimension reduction into a classifier model to train a classifier; and carrying out glutenite logging curve identification by utilizing the trained model. According to the method, the processing efficiency and the recognition precision of the logging curve data are effectively improved, the diversity and the quality of data generation are enhanced, the optimization problem in a traditional deep learning method is solved, and the generalization ability and the classification accuracy of the model for unknown data are further improved.