275 results about "Nonlinear transformation" patented technology

The invention provides a photovoltaic prediction method based on a data double decomposition and deep learning optimization model, and the method comprises the steps: collecting and preprocessing historical photovoltaic power data and meteorological associated data, carrying out the double decomposition of the preprocessed historical photovoltaic power data, and obtaining photovoltaic power component data; combining the photovoltaic power component data with meteorological associated data to construct a plurality of groups of photovoltaic-meteorological component data sets; an iTransform-KAN photovoltaic power prediction model is constructed, the photovoltaic-meteorological component data set is used to train and test the iTransform-KAN photovoltaic power prediction model, and the trained iTransform-KAN photovoltaic power prediction model is obtained; and determining a final predicted value through a linear superposition strategy based on the photovoltaic power component predicted value. According to the method, precise stripping of multi-scale features of photovoltaic power and adaptive learning of nonlinear transformation are realized, so that the capability of modeling a complex dynamic relationship is improved.

The invention provides a full-text retrieval method and system fusing various types of documents, and relates to the technical field of information retrieval, and the method comprises the following steps: obtaining document representation through document content extraction and structure recognition, generating a cross-modal semantic vector by using word embedding and nonlinear transformation, constructing a hierarchical index and a cross-document association graph, and obtaining a full-text retrieval result; the basic correlation score is calculated after the query request is received, and the comprehensive score of the candidate content segments is calculated based on the association graph to determine the optimal retrieval result, so that unified representation and retrieval of heterogeneous documents are realized, the cross-document retrieval precision and relevance are improved, and the processing capability of a retrieval system on complex queries is enhanced.

The invention discloses a PCB defect real-time detection method based on multi-scale feature fusion, and relates to the technical field of PCB defect real-time detection methods, and the method comprises the steps: obtaining a to-be-detected PCB image, carrying out the size normalization and pixel value standardization processing of the image, and obtaining a standardized image meeting the input requirements of a model; inputting the standardized image into a backbone network of a teacher detection model, and extracting a multi-scale primary feature map containing texture information in different directions through a grouping convolution structure; transmitting the multi-scale primary feature map to a neck network of a teacher detection model, and performing weighted fusion on feature maps of different scales by using a learnable weight to generate a multi-scale fusion feature map; and in an up-sampling path of the neck network, generating channel description information after global pooling is performed on the deep fusion feature map, generating a channel attention weight through nonlinear transformation, acting the weight on a primary feature map of a corresponding level, and outputting an enhanced feature map.

The invention belongs to the technical field of load prediction in a low-voltage distribution area, and particularly relates to a time sequence prediction method for attention mixed multi-scale decomposition, which comprises the following steps: S1, preprocessing original time sequence data to obtain a standardized sequence X; s2, inputting X into MJDA, and outputting uniform high-dimensional representation U after feature enhancement; s3, inputting U into TCDA, and carrying out cross-dimension dependence modeling and deep nonlinear transformation to obtain a final enhanced feature U1; s4, inputting U1 output by the TCDA into a hybrid expert predictor group; the predictor group is composed of K parallel expert predictors, and a corresponding expert prediction result is obtained; meanwhile, U output by the MJDA is processed through a noise perception gating network, and weight distribution U used for expert predictor fusion is generated; and according to the U, carrying out weighted summation on the output of the K expert predictors to obtain a prediction result. According to the method, high-precision and high-stability load prediction can be realized in a low-voltage distribution area environment with limited resources.

The invention provides a facial skin flaw enhancement method based on a Lab color space. The method comprises the following steps: firstly, acquiring an RGB face image and converting the RGB face image into a CIE Lab color space with uniform perception; then, performing differentiation treatment according to the manually selected skin flaw type: for the vascular flaw, extracting statistical characteristics of a component and driving adaptive nonlinear transformation, and generating a grey-scale map which highlights the red flaw; for pigment flaws, nonlinear transformation is carried out on the component L, and then collaborative linear weighting and feature amplification are carried out on the component L, the component a and the component b, so that a grey-scale map with highlighted pigment spots is generated. And finally, coloring the grey-scale map in the Lab color space through adjustable parameters to generate a high-contrast color enhanced image. The method overcomes the dependence on hardware and training data in the prior art, can clearly and adaptively enhance various flaws such as acnes, couperose streaks and color spots, shows robustness under different illumination, and can be widely applied to clinical beauty, later photography and real-time video processing.

The invention relates to an unmanned aerial vehicle trajectory tracking control method based on model prediction and preset performance constraint, and the method proposes to introduce Lyapunov stability constraint into a model prediction control framework and integrate a preset performance control mechanism, thereby achieving the unification of performance constraint and system stability analysis. Comprising the following steps: establishing a nonlinear system model based on unmanned aerial vehicle dynamics; position errors and attitude errors are defined, a preset performance function is constructed, and errors with performance constraints are converted into unconstrained errors through error normalization and nonlinear transformation; establishing a model prediction optimization problem on the premise of considering input saturation and stability constraints; designing an auxiliary control law based on the transformation error to construct a stability constraint; it is proved that the control strategy can ensure that errors meet preset performance constraints and system local asymptotic stability. According to the invention, stable and reliable trajectory tracking control of the unmanned aerial vehicle system can be realized, and the method has high tracking precision and good dynamic performance.

The invention discloses a multi-modal data joint embedding method based on a hierarchical progressive arithmetic interaction network, and the method comprises the following steps: obtaining text data and image data from the same semantic entity, and extracting a text local feature, a text global feature, an image local feature and an image global feature; inputting the text local feature and the image local feature into an atomic layer, and processing based on a Cartesian product to generate a first-order interaction feature; inputting the first-order interaction features into a combination layer, and carrying out nonlinear transformation processing to generate enhanced nonlinear interaction features; and inputting the nonlinear interaction features into the aggregation layer, and generating a multi-modal joint embedding vector in combination with the text global features and the image global features. The method effectively solves the problems of modal isomerism, single interaction level, lack of dynamic adaptability and the like in the prior art.

The invention discloses an electric power equipment safety long text knowledge retrieval method based on sparse and dense joint enhancement, which comprises the following steps of: encoding an electric power equipment safety long text and retrieval query by utilizing a pre-training language model for performing supervision and fine tuning on an electric power equipment safety field label data set, and generating a semantic word vector; a weighted sparse vector fusing semantic information and a dense semantic vector enhanced through nonlinear transformation are generated through fine-grained semantic enhancement branches; respectively calculating sparse similarity based on a common item weight product and dense similarity adopting a double-distance weighted fusion strategy combining a cosine distance and an Euclidean distance; and carrying out weighted summation on the two similarity scores through the learnable weight to obtain a final similarity, and realizing accurate sorting retrieval of the long text knowledge. The method gets rid of dependence on global features of sentence vectors, can accurately capture local key information in a long text, and has high retrieval precision and semantic robustness.

The invention discloses an electric power production management system security situation awareness method based on a cryptographic algorithm. According to the invention, by deeply fusing SM2, SM3, SM4 and other national cryptographic algorithms and power system characteristics, a full-link autonomous and controllable security protection system is constructed. In a data acquisition stage, SM4 encryption transmission and SM3 hash evidence storage are adopted to ensure confidentiality and integrity of data from a source to processing, and eavesdropping and tampering risks in a transmission process are effectively resisted; in a core situation assessment link, point multiplication operation of SM2 elliptic curve cryptography is innovatively introduced into a node aggregation process of a graph neural network, and graph structure mapping of physical topology of a power system is combined, so that the model can accurately capture implicit association and cascade influence between equipment, and the situation assessment accuracy is improved. At the same time, the recognition capability of the hidden attack mode is enhanced by using the nonlinear transformation of SM4, and the perception depth and anti-attack toughness of the system to the complex threats in the power production scene are improved from the bottom layer of the algorithm.

The invention discloses an image fusion method and system based on multi-semantic guidance and mixed experts. The problems that in the prior art, robustness is insufficient, and visual fidelity and semantic integrity are difficult to balance are effectively solved. According to the method, infrared and visible light images to be fused and task identifiers are obtained, and firstly, a CLIP network is utilized to extract high-level semantic features as guide vectors; and then inputting the image and the guide vector into a pre-trained hybrid expert (MoE) fusion network. According to the network, intermediate features are extracted through an encoder, a gating network dynamically activates part of expert subnets according to task identifiers and semantic vectors and calculates routing weights, adaptive nonlinear transformation and weighted fusion are carried out on the features, and finally a high-quality fusion image is reconstructed through a decoder. The method can adapt to different task requirements, and the calculation efficiency is remarkably improved while the image fusion quality and the semantic consistency are improved.

The invention belongs to the field of hyperspectral image classification, and discloses a hyperspectral image classification method and system based on multi-scale spatial-spectral joint representation and dynamic context modeling, and the method comprises the steps: carrying out the feature dimension reduction processing of a hyperspectral image through principal component analysis; spatial spectrum collaborative information of hyperspectral data is deeply mined through a multi-scale spatial spectrum joint characterization module, and adaptive fusion and enhancement of spatial spectrum characteristics under different scales are realized; a dynamic context modeling strategy is introduced, and the perception ability of the model to context information is optimized by establishing a long-range dependency relationship between features; advanced feature integration and nonlinear transformation are carried out through a multi-layer perceptron, and precise classification of hyperspectral image ground objects is completed. According to the method, the performance superior to that of a current mainstream method is obtained on three public data sets, and the effectiveness and generalization ability of the method are verified.

The invention discloses a mechanical arm state limited preset time point control method and system, and relates to the technical field of robot control, and the method comprises the steps: obtaining a joint position tracking error and a joint speed tracking error based on a joint motion state and an expected trajectory of a mechanical arm; based on constrained system states corresponding to the joint position tracking error and the joint speed tracking error, nonlinear transformation is carried out on the joint position tracking error and the joint speed tracking error through an unconstrained state function, and an equivalent unconstrained virtual error variable is constructed in combination with coordinate transformation; based on the unconstrained virtual error variable, designing a sectional sliding mode variable; and designing a second-order reaching law based on the sectional sliding mode variable to form a closed-loop dynamic state, and solving based on the closed-loop dynamic state to obtain a control torque so as to control the mechanical arm joint. According to the method, the trajectory tracking error of the mechanical arm can be converged at the moment specified by a user, the system state is ensured to meet strict physical constraints in the whole process, and the method can be applied to accurate trajectory control in a complex task scene.

The invention relates to a safety monitoring method for an irrigation and drainage project operation state and an intelligent system, and belongs to the technical field of artificial intelligence. The method comprises the following steps: collecting and marking irrigation and drainage project operation monitoring data through a sensor, and constructing a training data set; completing data normalization by combining quantile and median robust scaling with adaptive nonlinear transformation, and mining and screening high-order interaction features by combining a mutual information theory and a gradient boosting decision tree; constructing a deep classification network fusing physical prior and adaptive feature interaction, introducing physical constraint and multi-scale feature fusion, and optimizing a model through adaptive marginal classification loss and physical feature manifold alignment loss; real-time data is preprocessed and then input into the model, and operation state grade classification and graded alarm are achieved. According to the method, data noise can be inhibited, a multi-index coupling relationship can be mined, the interpretability and robustness of the model can be improved by integrating a physical rule, irrigation and drainage project abnormity can be accurately identified and early warned, and the method is suitable for intelligent safety monitoring of an irrigation area.

The invention provides a Markov jump-based mechanical arm preset time trajectory tracking control method, which comprises the following steps of: calculating a trajectory tracking error by establishing a two-degree-of-freedom mechanical arm dynamic model with a Markov jump characteristic; a trajectory tracking error is subjected to nonlinear transformation, a nonsingular terminal sliding mode surface is constructed in combination with a preset convergence time upper limit function, and then a control torque is synthesized through an equivalent control item and a robust switching item containing a sliding mode surface sign function and a switching gain to drive a mechanical arm. And the trajectory tracking error is difficult to converge in the preset time.

The invention discloses a complication prediction method and system based on an SHAP-random forest, and relates to the technical field of medical data mining. The method comprises the following steps: acquiring preoperative CT images and puncture path planning data, and calculating risk factors; performing numerical value standardization, nonlinear transformation and interactive feature generation on the risk factors to obtain feature vectors, calculating mutual information scores of feature values in the feature vectors, if the mutual information scores are greater than an experience threshold, retaining the feature values, and after traversal is finished, obtaining updated feature vectors; on the basis of a random forest model, taking the updated feature vector as an input value, and calculating a complication probability; quantizing the contribution degree of the characteristic value based on a Shapley value; according to clinical indexes, the risk threshold is dynamically corrected, the complication risk level is divided, and complication prediction is completed, the problems that a static threshold ignores the blood coagulation state difference of a patient and a black box model cannot provide a decision basis are solved, and the complication misjudgment probability is reduced.

The invention discloses a non-cooperative signal carrier frequency estimation method and system in a motion state, and relates to the field of wireless communication and signal processing, and the method comprises the following steps: selecting an energy prominent frequency point in a frequency spectrum of a non-cooperative signal, and combining the resolution of the frequency spectrum to obtain a carrier frequency coarse estimation value; carrying out frequency mixing down-sampling processing, windowing and nonlinear transformation on the non-cooperative signal, carrying out time-frequency conversion, and utilizing a DFT spectrum peak decimal correction method to obtain a frequency fine estimation value to compensate the carrier frequency coarse estimation value so as to obtain a carrier frequency fine estimation value; selecting an observation time period, and calculating a carrier phase observation value according to the carrier frequency fine estimation value; and after the carrier phase observation value is compensated according to the motion phase compensation item and is represented as a standard linear regression form, calculating the carrier frequency deviation to compensate the carrier frequency fine estimation value so as to obtain an accurate carrier frequency estimation value. According to the invention, the problem that the carrier frequency estimation deviation of the non-cooperative signal in the motion state is increased can be solved.

The invention provides a cross-platform public opinion information collection and analysis system based on deep learning, and belongs to the technical field of public opinion information collection and analysis. A multi-platform data collector is deployed to capture social media data in real time to establish an original cache pool, and preprocessing and metadata extraction are performed on multi-modal data; a multi-modal deep fusion convolution algorithm is adopted to construct a parallel text, image and audio branch network for feature extraction, a cross-modal attention mechanism is adopted to parallelly calculate correlation weights of all modal features, and a gating fusion unit is used to perform efficient nonlinear transformation and information screening to generate a unified semantic vector. Semantic vectors are input into a self-adaptive cross-modal sentiment analysis recognition model, and the technical problem that the requirement for millisecond-level real-time processing of mass public opinion data cannot be met due to the fact that the inference time of a deep model is too long is solved through a collaborative mechanism of dynamic optimization, parallel processing and hierarchical response.

The present disclosure relates to techniques for improving inference efficiency and memory utilization in transformer-based neural networks. A modified architectural design is introduced that decouples key and value matrix generation from inter-layer dependencies, enabling statically computed or parallelizable projections across layers. The disclosed approach may eliminate the need for layer-wise prefilling, support linear-time inference, and substantially reduce the memory footprint associated with key-value (KV) caching. The disclosed architecture may use shared or layer-specific projections, with a single KV-cache serving all or subsets of layers. In some embodiments, a non-linear transformation (e.g., implemented via a feed-forward network), may preprocess input embeddings prior to query, key, and value generation. A lookup table of transformed embeddings may be precomputed to further accelerate inference. The disclosed system can enhance scalability and may allow deployment of large models on resource-constrained hardware, offering practical benefits for latency-sensitive applications and long-context processing in transformer-based models.

The invention discloses an IV-type secretory effect protein recognition method and system based on multi-modal information, and belongs to the technical field of bioinformatics and secretory protein recognition. The method comprises the following steps: firstly, extracting amino acid residue characteristics of protein by using a protein language model, and constructing a spatial adjacency graph of the protein by using a three-dimensional structure predicted by a protein structure prediction model; respectively extracting sequence features and structural features of the protein through a deep sequence module and a hierarchical graph module; meanwhile, a contrast learning module is introduced to realize cross-modal alignment of the same protein in a potential space; and performing bidirectional interaction on the sequence features and the structural features by using a cross attention module to obtain joint features, and finally outputting a classification result after nonlinear transformation by a gating linear unit for predicting the IV-type secretion effect protein. According to the invention, efficient coordination of sequence and structure bimodal information is realized, and the identification accuracy of the IV-type secretory effect protein is remarkably improved.

The application discloses a light field image encoding method, comprising: constructing a multi-domain information coupled anchor point feature optimization module; taking a light field image and adaptively learned anchor point features as inputs, fusing spatial, angle and EPI information of the light field image with the adaptively learned anchor point features, and outputting anchor point features embedded with light field information; constructing an anchor point inter-context transformation module; taking center anchor point features embedded with light field information and features corresponding to the nearest neighbor anchor points as inputs, performing nonlinear transformation on the features corresponding to the anchor points, and outputting compact center anchor point features; and using an entropy model to compress and encode the center anchor point features; predicting the attributes of a Gaussian cell based on decoded anchor point features, and reconstructing a light field multi-view image in combination with volume rendering technology; constructing and training a 3DGS-based light field image encoding network; inputting an original light field multi-view image, and outputting a reconstructed light field multi-view image based on compact center anchor point features.

The invention relates to the technical field of gold mine filling automatic control, in particular to a gold mine filling automatic control method based on the Internet of Things. The method comprises the steps that original multi-modal data are collected and preprocessed, the preprocessed multi-modal data are subjected to tensor nonlinear transformation, and disturbance characteristic quantities are obtained; constructing a modal input vector based on the disturbance characteristic quantity, and processing the modal input vector to obtain a hidden layer embedding vector; based on the implicit layer embedding vector, carrying out implicit variable path evolution processing to obtain inter-modal path tension; constructing a control instruction based on the inter-modal path tension; after the control instruction is sent to the control equipment, original multi-modal data are collected in real time and predicted; and calculating a composite quantitative feedback error based on the multi-modal data predicted value, and updating and optimizing the control instruction. The problems that the filling quality control precision is low, the filling efficiency is not high, the manual dependence degree is large, filling parameter adjustment lags behind, and the sudden disturbance response capacity is poor in the current gold mine filling operation are solved.

Systems and methods are provided herein for advancing ensemble learning methods, including stacking, boosting, and bagging, for defeating data protection approaches by converting their generated unlearnable data into learnable ones. Processes of the present disclosure may enhance and implement ensemble learning on the unlearnable data while incorporating nonlinear transformations.

This application relates to a method and apparatus for determining process parameters in metal compound thin film deposition. The method constructs a two-stage cascaded process prediction model. The first stage, a feature extraction model, performs nonlinear transformations and feature extraction on the process parameters, extracting intermediate feature information that comprehensively characterizes the sputtering process state. The second stage, a probabilistic prediction model, establishes a probabilistic mapping relationship between the intermediate feature information and the thin film response information, decoupling the complex nonlinear coupling relationships between multiple parameters in reactive sputtering. It outputs predicted values ​​and corresponding confidence information. Based on the predicted values ​​and confidence information, process parameters are screened, considering both target performance requirements and prediction reliability requirements. This improves the efficiency and reliability of process parameter screening, reduces the number of invalid experiments, and enables the prediction of key response quantities such as thin film deposition rate and chemical composition even with limited experimental data, shortening the process development cycle.

The invention relates to the technical field of circuit design, in particular to a cascade cross array, a data processing method and electronic equipment, the array comprises at least one first row line, at least one second row line and at least one column line, and memristive devices are arranged at cross nodes of the first row line and the second row line and the column line; the voltage input module comprises at least one output end, the output end is connected with the first row line, and the output end applies input voltage to the first row line; a trans-impedance amplifier and an analog-to-digital converter are arranged on the second row line, the trans-impedance amplifier is used for clamping the voltage of the second row line to 0V and converting the current of the second row line into voltage, and the analog-to-digital converter outputs the voltage of the trans-impedance amplifier. Therefore, the problems of high power consumption, high delay, incapability of introducing nonlinear transformation in calculation and the like when a voltage input module, an analog-to-digital converter and a trans-impedance amplifier are used for many times are solved.

The application discloses the photovoltaic time sequence dynamic prediction method and system based on attention mechanism in the photovoltaic power generation prediction technical field. The method comprises the following steps: data preprocessing is carried out on the original photovoltaic time sequence data of a target region in a preset historical time period, and the preprocessed photovoltaic time sequence data is input into a trained photovoltaic time sequence dynamic prediction model; the preprocessed photovoltaic time sequence data is subjected to time sequence feature embedding to obtain an embedded feature vector; the embedded feature vector is input into an improved multi-head self-attention layer to perform dynamic attention calculation and competitive balance, and an optimized context feature vector is obtained; a feedforward neural network layer is used to perform nonlinear transformation and feature dimension integration on the optimized context feature vector, and a reinforced high-level feature vector is obtained; and an output layer is used to obtain a photovoltaic load prediction value sequence of a future specified time span. The application can improve the accuracy, stability and robustness of photovoltaic load prediction in a high-proportion photovoltaic access scenario.

The invention relates to the technical field of enterprise development evaluation, and particularly discloses an enterprise development potential evaluation method based on big data, and the method comprises the steps: collecting agricultural enterprise multi-source heterogeneous data through distributed nodes, importing a deep learning driven multi-modal fusion model, and generating a high-dimensional feature vector through cross-modal feature alignment and weighting; inputting the high-dimensional feature vector into a risk and pressure resistance evaluation model, constructing a dynamic risk conduction network mining risk association, generating a risk evaluation vector in combination with a time sequence feature, cascading the risk evaluation vector with the high-dimensional feature vector, and then generating a fusion feature vector through nonlinear transformation; and on the basis, a quantitative index is generated through multi-dimensional feature matching, and a comprehensive evaluation result is output through multi-scale decision analysis. According to the method, deep fusion of multi-source data is realized, risk changes are dynamically captured, evaluation comprehensiveness and accuracy are improved, and a reliable basis is provided for decision making of agricultural enterprises.

The invention relates to the field of image processing, in particular to a fruit disease detection method and system based on global memory and local contrast, and the method comprises the steps: obtaining to-be-processed features of a to-be-detected image of a fruit, carrying out the shuffling and dimensionality reduction of the to-be-processed features through a channel, and inputting the to-be-processed features into a plurality of cascaded global memory encoders; each global memory encoder obtains modulation features through a channel and a space attention mechanism in sequence, then generates first and second attention features through a window self-attention mechanism and an adaptive contrast perception gating attention mechanism, and outputs the first and second attention features through nonlinear transformation and feature integration after fusion; and identifying the output characteristics of the final-stage encoder to obtain the fruit disease condition. Compared with the prior art, the method has the advantages that local context dependence is modeled through the window self-attention mechanism, global historical information is modeled through the gating technology of the self-adaptive comparison perception gating attention mechanism, the local context dependence and the global historical information are fused for fruit disease recognition, and accuracy is greatly improved while reasoning efficiency is guaranteed.

This invention provides a method for modeling a cylinder motion system based on a physical information neural network, comprising: acquiring time-series data of the cylinder motion system, including time-series control quantities and system state quantities; constructing a physical information neural network model with a dynamic feature input layer and a physical parameter input layer, receiving the time-series control quantities and physical parameter vectors respectively; performing a nonlinear transformation on the physical parameter vectors through a physical parameter encoding network to obtain encoded physical feature vectors; fusing the time-series control quantities and physical feature vectors to form a joint input vector; inputting the joint input vector into the backbone neural network, synchronously predicting the system state quantities through multi-layer nonlinear transformations; constructing a composite loss function including a data loss term and a physical loss term; training the model using time-series data, updating the model parameters by minimizing the composite loss function until the model converges, thereby achieving high-precision modeling using only single motion data.

The invention provides a copper-based composite material performance prediction method based on a space-time attention mechanism, and the method comprises the steps: obtaining a microstructure diagram and stress-strain data through molecular dynamics simulation, extracting topological features through a diagram attention network, and dynamically adjusting the parameters of a time sequence convolution network; a cross-space-time attention module is constructed, and bidirectional feedback fusion of space feature screening and time sequence feature optimization is realized; and finally, carrying out dimensionality reduction and nonlinear transformation on the fused features through a full-connection layer, and outputting performance prediction values such as yield strength and Young modulus. According to the method, the precision and reliability of performance prediction of the copper-based composite material can be improved, and efficient and accurate theoretical support is provided for material design and performance optimization.

The application relates to the field of image processing, in particular to a fruit disease detection method and system based on global memory and local contrast. The method comprises the following steps: obtaining to-be-processed features of a fruit to-be-detected image, inputting the to-be-processed features into a plurality of cascaded global memory encoders after channel mixing and dimension reduction; each global memory encoder obtains modulated features through channel and spatial attention mechanism in turn, then generates first and second attention features through window self-attention mechanism and adaptive contrast perception gated attention mechanism respectively, and outputs after fusion, nonlinear transformation and feature integration; and the output features of the last-stage encoder are recognized to obtain the fruit disease condition. Compared with the prior art, the local context dependence is modeled through the window self-attention mechanism, the global historical information is modeled through the gating technology of the adaptive contrast perception gated attention mechanism, and the fruit disease is identified by fusing the two, so that the accuracy is greatly improved while the reasoning efficiency is ensured.