334results about How to "Improve discrimination ability" patented technology

The invention provides a three-dimensional point cloud data segmentation method and system based on multi-scale point features, and relates to the technical field of point cloud processing, and the method comprises the steps: obtaining original point cloud data, constructing a three-dimensional Hash index table, carrying out the self-adaptive hierarchical division through a local maximum value of a density distribution curve, and obtaining a local maximum value of a density distribution curve; noise point elimination is carried out based on a region density count value and a geometric distribution coefficient, an enhanced feature vector containing a main feature direction, a main curvature value and the like is constructed for non-eliminated points, a local geometric descriptor is constructed by adopting four spherical search regions with increasing radiuses, a three-branch space attention network is input to extract spatial features, and the spatial features are extracted; and fused feature data are obtained through a multi-level geometric feature extraction network and a bidirectional feature propagation network. And obtaining a final segmentation result through segmentation energy function optimization and a region merging strategy.

The invention relates to a multi-modal implicit sentiment analysis method based on a multi-granularity attention mechanism, which belongs to the field of artificial intelligence, and comprises the following steps: S1, extracting visual features in image information by using a visual feature extraction strategy of multi-scale attention fusion; s2, generating a text description of the image information by adopting an image translation engine, splicing the text description with an original text, and extracting text features through a hierarchical attention mechanism; s3, carrying out normalization processing; s4, performing multi-modal fine-grained correlation fusion on the visual features and the text features by adopting a bidirectional cross attention mechanism, and extracting fine-grained features in the visual features and the text features through a multi-head attention mechanism and residual connection; and S5, fusing the fine-grained features by using an adaptive multi-modal classification module to complete sentiment classification. The multi-modal sentiment analysis method solves the problems that multi-modal sentiment analysis is insufficient in inter-modal correlation mining, insufficient in fine-grained information capture and large in noise interference.

The invention provides a remote sensing image target detection method based on an improved FCOS, and relates to the technical field of target detection, and the method comprises the following steps: carrying out the convolution and pooling operation of a remote sensing target image based on a pre-trained convolutional neural network, generating a preliminary feature, and carrying out the enhancement processing of the preliminary feature through a twin multilayer perceptron; carrying out fusion processing on different levels of feature information in the multi-level feature map through a feature map splicing mode and a feature map enhancement mode; and determining the position of the remote sensing image target according to the analysis result to obtain a target detection result. According to the method, the feature pyramid network is combined with an innovative coordinate attention mechanism, fusion of high-level semantic information and low-level detail information is realized, an enhanced multi-scale feature map is generated to cope with diversity of target scales in a remote sensing image, multi-scale fusion and an anchor-frame-free design are combined, false detection is effectively reduced, and the detection accuracy is improved. Therefore, the method shows efficient, accurate and wide application potential in the field of remote sensing image target detection.

The invention relates to the technical field of intelligent lock recognition, in particular to an intelligent lock image feature recognition method and system based on deep learning. The recognition method comprises the following steps: constructing a multi-branch deep neural network model, wherein the multi-branch deep neural network model comprises a backbone network, a feature extraction branch and a feature fusion branch; a framework training model is used to learn identity recognition, attitude estimation and environment perception tasks; and identifying the comprehensive feature representation by using a trained framework training model, and generating a control instruction of the intelligent lock according to an identification result. According to the intelligent lock image feature recognition method based on deep learning provided by the invention, comprehensive analysis of multi-dimensional information such as identity, posture and environment is realized by constructing the multi-branch deep neural network model.

The invention provides a traditional Chinese medicine abnormal state recognition method, system and device and a storage medium, and relates to the field of abnormal recognition. The method comprises the following steps: acquiring product information of a to-be-detected traditional Chinese medicine, and determining a data acquisition standard of the to-be-detected traditional Chinese medicine based on the product information of the to-be-detected traditional Chinese medicine and a preset detection rule; based on a data acquisition standard of the traditional Chinese medicine to be detected, inputting the image data into a preset image anomaly detection model to obtain a preliminary image anomaly region; performing feature fusion on the visual feature data and the spectral feature data to obtain an abnormal feature vector; and inputting the abnormal feature vector into a preset abnormal state judgment model, outputting abnormal state data, and performing abnormal state judgment based on the abnormal state data and a preset abnormal state judgment threshold to obtain an abnormal state detection result of the to-be-detected traditional Chinese medicine. Through the scheme, the abnormal state of the traditional Chinese medicine can be identified timely and accurately.

The invention provides a medical image classification unsupervised domain self-adaption and multi-granularity alignment system and a medical image classification unsupervised domain self-adaption and multi-granularity alignment method. According to the method, a coarse-to-fine progressive feature optimization strategy is provided, that is, a fine-grained feature encoder is initialized by using the training weight of a coarse-grained feature encoder, and alignment optimization of fine-grained features is guided while original semantic information is reserved. Specifically, a staged and layer-by-layer unfreezing fine adjustment method is adopted, and the adaptive capacity of fine-grained features is enhanced on the premise that the invariance of a coarse-grained domain is not damaged. According to the strategy, the generalization and convergence efficiency of the model on target domain data is remarkably improved. And experimental verification is carried out on a plurality of cross-domain medical image classification data sets. Results show that no matter on a small-scale data set or a large-scale data set, the method provided by the invention can obtain performance obviously superior to that of an existing method, and generalization ability and robustness are stronger.

The invention discloses a coronary heart disease preliminary screening method and device based on facial feature analysis and a medium, and relates to the technical field of image understanding, and the method comprises the steps: extracting each part sub-region of a human face, and obtaining a tensor image of each sample data through the stacking of the sub-regions; obtaining the flattened feature maps of the tensor images, and integrating the feature maps to obtain a feature matrix of the corresponding sample data; calculating a feature mean value in the image dimension direction according to the feature matrix, and taking the feature mean value as a global feature tensor of a corresponding sample; carrying out dimension reduction processing and nonlinear transformation on the global feature tensor, and carrying out binary classification probability distribution output of relevance between the subregions and the coronary heart disease; performing updating iteration of the classification weight of the sub-region through cross entropy loss function constraint and a stochastic gradient descent optimizer; and performing primary screening of the coronary heart disease through the iterated primary screening model of the coronary heart disease. According to the method, the key feature capturing capability of the model is improved, and then the judgment capability on the coronary heart disease risk is enhanced.

The invention relates to the technical field of big data processing, in particular to an online state grid user classification method, system and device based on a graph convolutional neural network and a storage medium. The invention discloses an online state grid user classification method based on a graph convolutional neural network. The method comprises the following steps: extracting user explicit and implicit features; the method specifically comprises the following steps: firstly, constructing a user graph model by taking online state grid user data as input, extracting feature information in graph data by using a stack network formed by a convolutional neural network, and propagating vector representations of aggregated users and neighbor users thereof by using a cyclic convolution operation; classifying users; according to the method, the graph convolutional neural network model is used for constructing the portrait of the state grid user on the Internet, dominant and implicit features of the user are deeply mined, and an accurate user feature vector is formed. Meanwhile, user vector subdivision is achieved according to probability representation output by the full connection layer, and potential user groups are effectively recognized.

The invention discloses a speech emotion recognition method based on multi-modal feature fusion, and the method comprises the steps: designing a multi-modal speech emotion recognition model, enabling an audio to serve as a main mode, employing the function of wav2vec2.0, and combining a text mode with a visual mode to enhance the performance of emotion recognition based on the audio; the self-attention mechanism, the GRU and the LVC are combined to extract global and local information from a visual mode, the global information is extracted only from text features, visual clues and text clues related to emotion can be comprehensively understood, the extracted text and visual features are integrated into a wav2vec2.0 model, multi-mode representation of multi-level audio information is formed, and the multi-mode representation of the multi-level audio information is realized. And meanwhile, the discrimination capability of the model is enhanced by utilizing multi-modal homogeneous feature difference learning, and the proposed model is evaluated on two data sets, so that the model is proved to have excellent robustness and generalization compared with the existing most advanced model.

The invention relates to the technical field of medical information, and discloses a pancreatic cancer risk prediction method based on machine learning and multi-modal data, and the method comprises the steps: obtaining the multi-modal data of a target user; performing time sequence deduction on the molecular biological detection data to generate a virtual molecular time sequence; time sequence signals are extracted from the virtual molecule time sequence and the time sequence behavior monitoring data; calculating the dynamic coupling strength between the two time sequence signals to obtain a space-time coupling coefficient; weighted fusion is carried out on the features, and unified multi-modal feature representation is constructed; carrying out multi-modal feature representation training to obtain a special risk prediction model for the target user; and obtaining a risk quantitative score, and identifying a key risk driving factor which contributes to the score most. According to the invention, through multi-modal time sequence fusion and personalized modeling, early-stage, dynamic and explainable and evaluable pancreatic cancer risks are realized.

A method, system and device for monitoring downhole flow and moisture content specifically comprises the steps that S1, DAS signals are collected, and the obtained DAS signals are preprocessed; s2, signal features in the DAS are extracted, and time domain features, frequency domain features, time-frequency domain features and time sequence features in the DAS are extracted; s3, the DAS signal features extracted in the S2 are fused, and comprehensive feature representation is generated; and S4, identifying the DAS signal features, training an AI model to obtain a flow water content identification model, and identifying the comprehensive feature representation through the flow water content identification model to obtain flow and water content identification. The method has the advantages that four DAS signal feature extraction methods are provided, and time-frequency feature extraction, frequency-frequency feature extraction, time-frequency feature extraction and time sequence feature extraction are achieved respectively; designing a feature fusion algorithm, and realizing feature-based event discrimination capability enhancement; a deep learning algorithm implementation method is provided, and a high-precision feature recognition technology is improved.

The invention relates to the technical field of computer vision, and discloses a visual saliency target detection method based on an RGB-D image, and the method comprises the following steps: S1, carrying out the channel alignment and spatial calibration of a mutual relation between corresponding features of different modals through an MI module, and obtaining a cross-modal representation; s2, then, an IGF module is added to the tail end of the Swin Transform encoder, and the IGF module is added to the tail end of the Swin Transform encoder; s3, the cross-modal fusion feature # imgabs0 # is subjected to channel cross fusion through a CCT module, and relevant information about significance is further obtained; and S4, finally, carrying out modeling on the remote dependency of the fusion features from a global perspective by using a decoding block based on Swin-Transform. According to the visual saliency target detection method based on the RGB-D image, by designing a modal interaction module and a channel cross fusion mechanism, the fusion effect of multi-modal features and the fine modeling capability of multi-scale features are fundamentally improved, and therefore the problems of saliency target detection precision and robustness which are difficult to overcome in the prior art are effectively solved.

The invention discloses a multi-modal false information detection method, system and device and a storage medium, and belongs to the technical field of false information detection.The method comprises the steps that a text of multi-modal information is input into a text representation network to obtain text representation, and an image of the multi-modal information is input into an image representation network to obtain image representation; and inputting the text representation and the image representation into a shared classifier to obtain a text prediction information label, an image prediction information label, a text weight and an image weight, and weighting the text prediction information label and the image prediction information label based on the text weight and the image weight to obtain a prediction information label. According to the method, the text and image modal information is combined, and an adaptive weight distribution mechanism is adopted, so that efficient and accurate multi-modal information processing is realized, the prediction performance is remarkably improved, and the application range is widened.

The invention belongs to the field of multi-label image classification, and particularly discloses a neural collapse-based long-tail multi-label image classification method, which guides learning of different types of features by using label embedding of an equiangular tight frame, so that the different types of features meet the property of neural collapse, the difference between different types is maximized, and the neural collapse-based long-tail multi-label image classification efficiency is improved. The influence of the head class on the tail class in the same sample is reduced, and learning of tail class features is fully guaranteed. Besides, a comparative learning method is also adopted, so that each class feature is clustered and collapsed to a class prototype, and the model learns more robust features. And finally, features corresponding to different classes of samples extracted by the model are combined and input into a binary classifier meeting an equiangular tight frame, so that the sample features are fully utilized, the influence of an unbalanced data set on the training of the model is avoided, the generalization ability of the model is enhanced, and the training efficiency of the model is improved. And the prediction precision of multi-mark image classification is improved.

The invention discloses a hyperspectral image classification method based on frequency domain denoising and element gradient correction, and the method comprises the following steps: carrying out the preprocessing of all hyperspectral image data, and dividing an overall training sample set formed by the processed hyperspectral images into a training set and a verification set; constructing a sample weighting model based on frequency domain denoising and element gradient correction; in the training process, a parameterization frequency spectrum gating sensing transformation module is utilized to map features to a frequency domain through discrete Fourier transform, a learnable frequency spectrum response function is utilized to adaptively suppress spectrum jitter noise, and finally pure features are reconstructed. And automatically constructing a high-confidence pseudo-clean verification set based on a Gaussian mixture model and time domain consistency. According to the method, a time domain momentum updating mechanism is introduced, the variance of statistical estimation is effectively smoothed, random interference caused by training fluctuation is resisted, and the accuracy of pseudo clean set construction and the convergence stability of overall model training are further improved.

The invention discloses a multi-task electromagnetic model based on a hybrid expert network, and belongs to a wireless communication technology. The model comprises a preprocessing module, a feature extraction module, a task output module and a pre-training-fine tuning learning strategy. The preprocessing module carries out standardization processing on the multi-source electromagnetic signals; the feature extraction module is based on a Transform structure, introduces a hybrid expert network to replace part of a traditional feedforward neural network, and dynamically selects an expert sub-network through a task specific routing mechanism; the task output module configures a special structure according to different task targets; in the pre-training stage, a mask auto-encoder is used for pre-training large-scale label-free data, and a downstream task is subjected to full-amount fine adjustment through small-scale label data. According to the method, multi-task collaborative learning and differential expression are realized, and the recognition performance, robustness and processing efficiency of the model in a complex electromagnetic environment are improved.

The invention relates to the technical field of semantic segmentation, in particular to a night semantic segmentation method based on low illumination enhancement and edge optimization, and the method comprises the steps: inputting an image into a low-light enhancement repair network based on the Retinex theory, local contrast enhancement and adaptive feature fusion, obtaining a denoised and enhanced intermediate image, and carrying out the edge optimization of the intermediate image; inputting the intermediate image into a semantic segmentation network to obtain a category distribution diagram of each pixel; inputting a discriminator embedded with a channel attention module according to the category distribution diagram of each pixel, and optimizing a generator composed of a low light enhancement repair network and a semantic segmentation network through a multi-task joint optimization loss function; and inputting a night image to be detected and segmented into the optimized generator, and outputting a segmentation result. By adopting the method, the low-light enhancement repair network is combined with a local contrast enhancement and channel feature fusion mechanism, so that the overall brightness of the image is improved, the details and edge structures of the image are reserved, and the perception capability and robustness of the model are improved.

The invention discloses an ICBL-based intelligent fault diagnosis method for a gear box of a cutting unit of a coal mining machine. The method comprises the following steps: firstly, collecting vibration data of different fault types; randomly dividing the data into a training data set and a test data set, and setting an unbalance degree in the training data set; then, complex structure information of the vibration data of the gearbox of the cutting part of the coal mining machine can be quickly and efficiently embedded into the improved cascade width learning model by using a random hypergraph convolution mechanism, so that the discrimination of fault features learned by the improved cascade width learning model is greatly improved; meanwhile, a class specificity weight distribution strategy is designed according to proportion information between data classes, an improved cascade width learning model can pay more attention to a few class samples, the fault diagnosis performance of unbalanced data is improved, and the fault diagnosis efficiency of the unbalanced data is improved through a mode of combining a random hypergraph convolution mechanism and the class specificity weight distribution strategy. And finally, the fault diagnosis precision of the gear box of the cutting part of the coal mining machine is effectively improved.

The invention relates to the technical field of electric data processing, in particular to a data blood relationship automatic tracking method and system based on a table in AI data, and the method comprises the steps: obtaining metadata in the table in the AI data, and analyzing a task log of a calculation engine to construct an initial directed acyclic graph; the initial directed acyclic graph comprises data nodes and explicit blood edges connected with the data nodes; and calculating a flow entropy coupling index between any two data nodes. According to the method, a flow entropy coupling index based on update time correlation and logic evolution divergence based on key entity distribution similarity are constructed, a graph comparison learning model is combined, under the condition that an explicit SQL grammatical structure is lacked, the implicit dependency of upstream and downstream nodes is automatically identified, and the flow entropy coupling index and the logic evolution divergence are combined. And end-to-end full-link data consanguinity complete tracking in a complex AI feature engineering scene is successfully realized.

The invention relates to the technical field of AI visual detection, and discloses a glass fiber filament flying and breaking early warning and disposal method based on AI visual analysis, which comprises the following steps: collecting a normal operation video of a coil base, processing the video according to frames to form a plurality of pictures, manually marking the pictures containing broken filaments, and training a model by using the marked pictures to obtain a broken filament early warning and disposal result; the coil base is detected in real time through the model, when it is detected that the number of broken wires reaches a threshold value, the audible and visual alarm of the coil base where the broken wires are located is started, alarm information of the coil base where the broken wires are located is obtained, the alarm information is processed through a definition rule, a hexadecimal instruction is generated, and the instruction is sent to a corresponding worker. And meanwhile, the work number information of the worker is judged through an optical character recognition model.

The invention discloses a tailing dry beach automatic detection and early warning method based on multi-sensor image combined processing, and belongs to the field of image data processing. According to the method, an unmanned aerial vehicle carrying an RGB camera, a multispectral camera and a laser radar is used for collecting RGB images, multispectral images and three-dimensional point cloud data, then various feature data are obtained, boundary information is obtained through fusion and a random forest classifier, then the length of a dry beach is calculated, and early warning is carried out. According to the method, the prediction accuracy and efficiency are improved, the advantages of various data are fully utilized, the tailings pond monitoring has higher reliability and flexibility, and meanwhile the method has the advantages of being good in random forest classification effect, high in discrimination capability, low in false drop rate and the like.

The invention discloses a multi-modal contrast learning fault diagnosis method for a small sample scene, and the method comprises the following steps: carrying out the enhancement of a one-dimensional signal and two-dimensional image fused fault data set through physical simulation for the small sample scene with scarce industrial fault data, and constructing a positive and negative sample pair through a plurality of data enhancement strategies; based on heterogeneous multi-modal fault data, designing a double-flow encoder architecture of a time sequence branch and an image branch, extracting depth features and mapping the depth features to a unified feature space through a projection head; performing supervised contrast learning pre-training based on intra-modal and inter-modal dual contrast loss; supervision fine tuning is carried out based on multiple loss functions such as physical guidance, so that accurate diagnosis of equipment faults is realized in a small sample scene. According to the fault diagnosis method under the unbalanced sample and limited labeling conditions, the problem that a traditional data driving model depends on large-scale labeling samples is effectively relieved through supervised comparative learning and cross-modal information alignment.

The invention relates to a construction method and device of a substation anomaly detection network, equipment and a storage medium. The method comprises the following steps: acquiring a plurality of abnormal image samples of substation equipment and an abnormal label corresponding to each abnormal image sample; performing frequency domain feature extraction on the target abnormal image sample by adopting a to-be-trained substation anomaly detection network to obtain a frequency domain feature, and performing classification based on the frequency domain feature to obtain a predicted anomaly corresponding to the target abnormal image sample; calculating model loss based on the abnormal label corresponding to the target abnormal image sample and the prediction abnormality; adjusting network parameters of the to-be-trained substation anomaly detection network based on the model loss until the calculated model loss reaches a training stop condition, and obtaining a trained substation anomaly detection network; and the trained substation anomaly detection network is used for substation anomaly detection. By adopting the method, the detection precision of the multi-classification anomaly in the substation environment can be improved.

The invention relates to a BERT and multi-module fusion-based track circuit fault label classification method and device and a medium, and the method comprises the steps: obtaining a to-be-classified track circuit fault text, and carrying out the cleaning and standardization processing of the to-be-classified track circuit fault text; performing word segmentation and vectorization processing on the track circuit fault text to generate a lexical symbol sequence conforming to BERT encoder input specifications; extracting a context semantic vector representation of the input lexical symbol sequence through a BERT encoder; performing feature extraction through a TextCNN module and a self-attention module which are parallel to obtain local features and weighted global features, and splicing the local features and the weighted global features to obtain a comprehensive feature vector; and synchronously generating a multi-label classification result of a fault phenomenon, a fault reason and a solution measure corresponding to the current fault text through a multi-label classification head based on the comprehensive feature vector. Compared with the prior art, the method has the advantages of high discrimination, high accuracy, high practicability and the like.

The invention discloses a police target tracking system based on deep learning, and relates to the field of police video monitoring. The system comprises a video sensing and front-end processing unit, an edge intelligent tracking server and a cloud model optimization and command platform, the video sensing and front-end processing unit collects visible light and thermal infrared video streams, the edge intelligent tracking server processes the visible light and thermal infrared video streams, and a stable track with an identity label is generated; and the cloud model optimization and command platform aggregates edge data based on a federated learning framework, iteratively optimizes a global model and realizes command scheduling. According to the system, the tracking robustness, the real-time performance and the multi-target identity keeping accuracy in a complex environment are effectively improved, and the actual combat requirements of police affairs are met.

The invention discloses a cross-city traffic prediction method and system based on multi-modal fusion and spatial expert routing, relates to the technical field of traffic prediction, and provides an adaptive modal selection mechanism based on a signal-to-noise ratio for the problems of missing, noise and uneven quality of multi-modal data in different cities. And a low-quality mode is dynamically suppressed in combination with comparative learning, and robust multi-mode fusion is realized. In order to solve the problems of large space structure difference and weak generalization ability among cities, a multi-modal guided space expert routing architecture is designed: modal sharing experts and routing experts are activated by using a multi-modal context, and local space dependence is adaptively modeled for different functional regions. The method supports any modal combination input, city specific fine tuning is not needed, and the zero sample cross-city prediction performance is significantly improved.

The invention relates to the technical field of clinical medical treatment, and discloses a disease deterioration risk identification method and system based on depth time sequence clustering enhancement, and the method comprises the steps: 1, obtaining multi-modal clinical sequence data of a patient, including physiological indexes of a time sequence, a laboratory detection result, historical diseases and medication data; 2, performing feature extraction and classification on the patient sequences by adopting a knowledge enhanced sequence clustering method, and grouping the patient sequences according to future outcome distribution of the patient sequences; 3, enabling the model to quickly adapt to a prediction task of a new patient subgroup through a meta-training process; and step 4, based on the trained meta-model, carrying out rapid adaptation on the new patient subtype, and predicting the possibility that the new patient subtype has a deterioration event in a certain time window in the future. The method and the system can effectively learn the disease change mode of the patient under the condition of limited clinical data, improve the prediction accuracy of the new patient subgroup, and are especially suitable for clinical prediction scenes under the condition of small samples.

The invention discloses a multi-frequency sound wave analysis method based on reinforced concrete damage detection, and particularly relates to the technical field of multi-frequency sound wave analysis, and the method specifically comprises the steps: collecting multi-frequency emission information according to emitted multi-frequency sound waves, obtaining a frequency vacancy coefficient according to the multi-frequency emission information, collecting multi-frequency receiving information according to the received multi-frequency sound waves, and obtaining a frequency vacancy coefficient according to the frequency vacancy coefficient; a multi-frequency fusion coefficient is obtained according to the multi-frequency receiving information, a loss model is constructed according to the multi-frequency fusion coefficient and a frequency vacancy coefficient, a space loss index of the loss model is calculated, the space loss index is compared with a preset space loss threshold value, and an identification signal is generated according to a comparison result; an anomaly detection model is established according to the generated identification signal, classification evaluation is performed by using space loss index data of an abnormal state, and an intervention strategy is proposed according to a classification evaluation result, so that the problem of abnormal reinforced concrete damage detection fluctuation caused by frequency-by-frequency emission of multi-frequency signals is effectively solved; and the effectiveness and credibility of reinforced concrete damage detection are improved.

The invention discloses a visual box stacking method for a gantry crane, and relates to the field of automatic box stacking, and the method comprises the following steps: step 1, mounting a plurality of cameras on a lifting appliance, and enabling the collection angles of the cameras to cover the range of the lifting appliance; 2, initial lifting appliance working pictures are collected through a camera, a distortion line segment on the collected pictures is marked, and camera distortion parameter calibration is carried out; step 3, performing camera radial distortion removal on the originally collected picture through the calibrated distortion parameter to obtain a camera radial distortion removal picture; through effective camera calibration and distortion correction, the precision and consistency of image acquisition are ensured, the container identification accuracy is further improved, the condition that the image acquired by the camera is closer to the real world is ensured, the discrimination capability of the model on the position and attitude of the container is improved, and the identification accuracy of the container is improved through real-time monitoring and data analysis. Fine adjustment control can be rapidly carried out on the lifting appliance, and it is ensured that all parameters in the box stacking process are kept within a safe and effective range.

The invention provides a redried tobacco leaf mildewing detection method based on mixed attention mechanism enhancement and a 3D residual CNN network, and the method comprises the steps: collecting redried tobacco leaf samples, and preparing mildewed tobacco leaf samples in different periods; hyper-spectral image data of the mildewed and redried tobacco leaf samples are collected through hyper-spectral image collecting equipment; the collected hyperspectral image data is corrected; establishing a mildewed tobacco leaf data set according to the corrected hyperspectral image data; an MA3D-ResNet model is constructed; carrying out iterative training on the model by utilizing the training set and the verification set, and storing model parameters with the highest accuracy on the verification set within a specified number of iterations; model parameters which are obtained through training and have the highest accuracy rate are imported into the model, test set data are input, evaluation indexes are calculated according to a model output result, the corrected hyperspectral image of the redried tobacco leaves to be detected is input into the tested model, and a mildew detection result is obtained. The method can effectively improve the detection precision of the mildewing degree of tobacco leaves with extremely similar spectral characteristics, especially early mildewing tobacco leaves.