21results about How to "Reduce overfitting" patented technology

This invention discloses a bearing remaining life prediction method based on frequency domain degradation perception. It constructs an integrated prediction system through a time-frequency decomposition dual-branch feature extraction module, a frequency domain degradation perception weight allocation module, and a life prediction loss function optimization module. First, the time-frequency decomposition dual-branch feature extraction module decomposes the original time series into high and low frequencies and inputs it into an sLSTM and mLSTM dual-branch structure for feature extraction. Second, the frequency domain degradation perception weight allocation module performs degradation perception processing on the fused features, generating dynamic weights and modulating the features to obtain the final fused features, which are then mapped to the prediction space. Finally, the life prediction loss function optimization module calculates the loss, optimizes the model output, and obtains the bearing remaining life prediction result. This invention comprehensively covers the full-stage features of bearings from healthy to severely degraded, improves the sensitivity of early fault detection, and still possesses excellent generalization and robustness under complex operating conditions.

The application discloses a cross-natural language code retrieval model training method, a cross-natural language code retrieval method, a device, equipment and a medium, and relates to the technical field of artificial intelligence and software engineering. The cross-natural language code retrieval model training method comprises the following steps: obtaining an original corpus database, and constructing training data according to the original corpus database; performing confusion and inversion on main language codes to obtain main language code samples, wherein the main language code samples comprise main language code positive samples and main language code negative samples; and training an initial model through a gradient inversion layer according to the training data and the main language code samples to obtain a target model. According to the application, the natural language-specific "fingerprint" features in the codes can be removed, the embedding space alignment direction can be unified, the sampling distribution deviation in the training process can be reduced, and the consistency and generalization capability of cross-language code retrieval can be improved.

The present application relates to the technical field of ship identification detection, and more particularly to a layered attention fusion and edge enhancement SAR ship detection method and system. The method comprises the following steps: obtaining initial ship data images and preprocessing to obtain input feature maps; constructing an enhanced image recognition model, training the enhanced image recognition model according to the input feature maps to obtain a trained image recognition model; the enhanced image recognition model comprises a multi-scale edge information selection module, a cross-domain feature gating module and a layered attention fusion block module; and inputting an actual SAR image into the trained image recognition model to obtain a ship detection result. By designing and using the multi-scale edge information selection module, the cross-domain feature gating module and the layered attention fusion block module, the present application realizes better training stability and generalization ability in a few-shot detection scenario, effectively alleviating the overfitting problem of the baseline model when data is scarce.

The application discloses a face forgery detection method and system based on federal incremental learning, comprising the following steps: constructing a federal incremental learning framework, each client is provided with a local face forgery detection model, and a global server is provided with a global face forgery detection model; the local face forgery detection model is subjected to cross-entropy training in a basic training stage, generates a forgery substitute sample in an incremental training stage, and is used as training input together with a real face sample and a forged face sample; after training, the model parameters are uploaded to the global server for aggregation; the global face forgery detection model is updated based on a difference perception aggregation strategy; an adversarial perturbation is trained based on the global face forgery detection model converged in the present stage and the real face sample, and a perturbation pool is refreshed until a preset termination condition is reached; and a to-be-tested face image is identified as real or fake based on the trained global face forgery detection model. The application can relieve catastrophic forgetting and improve cross-task generalization ability and robustness.

The application provides a multi-branch graph adaptive network for individualized motor imagery electroencephalogram signal classification and a method thereof. Through an adaptive matching technology, a graph convolutional neural network (GCN) is used to learn a mapping relationship between an EEG signal and an optimal time-frequency domain processing method, and an optimal processing method is matched for each individual. The multi-branch network comprises an original data branch, a CWT branch and an STFT branch, and corresponding feature extraction networks are respectively designed. A data enhancement method based on a super-resolution generative adversarial network is used to expand a training data set and improve the generalization ability of the model. Experimental results show that the method significantly improves the classification accuracy of individualized EEG signals and has practical application value.

The present disclosure provides a training method of an image extraction model and an image extraction method, which can be applied to the fields of image processing and pattern recognition. The image extraction model comprises a linear embedding network, an encoding network and a decoding network. The training method comprises: processing a sample image obtained by using the linear embedding network to obtain an embedding feature map; processing the embedding feature map by using the encoding network to obtain i encoding feature maps; processing the i encoding feature maps by using the decoding network to obtain a weight fusion feature map, wherein the decoding sub-network is constructed based on a local-global attention layer and a weight feature fusion layer; performing segmentation head mapping processing on the weight fusion feature map to obtain an image segmentation result, wherein the image segmentation result represents a geological change attribute of a target geographical environment region; and training the image extraction model according to the image segmentation result and label data corresponding to the image segmentation result to obtain a trained image extraction model.

This invention discloses a computer program product and its application for data processing devices in the field of bioinformatics for ctDNA variant detection. The technical problem this invention aims to solve is how to detect ctDNA variants in early-stage cancer or postoperative minimal residual disease (MRD) under conditions of no UMI library construction and moderate sequencing depth. This invention constructs a set of supporting sequences for candidate variant sites, generating a sequence feature tensor and a fragment physical feature vector (including normalized fragment length). The former is input into a first neural network branch to extract sequence representation, and the latter into a second neural network branch to extract physical representation. A fusion module combines the sequence representation and physical representation, and a gating unit calculates the gating weight based on the physical representation and dynamically adjusts the contribution of the sequence representation, outputting the probability of the true ctDNA variant. This invention utilizes physical laws to suppress sequencing noise and can be applied to monitor MRD under conditions without molecular barcodes.

The application discloses a county scale soybean yield prediction method based on reference year time series difference modeling, and relates to the technical field of agricultural remote sensing and intelligent information processing. The method first acquires multi-source crop growth data and administrative level soybean yield data of previous years in a research area, and constructs a spatial scale multi-source time sequence characteristic sequence of a unified space-time scale after preprocessing; then, a reference year is selected to construct a sample pair, and a multi-source time series difference sequence and a yield difference value are calculated; subsequently, a deep learning model is constructed, the difference sequence is taken as input, and the yield difference value is taken as a label to train the model; finally, the difference sequence of a to-be-measured sample is input to obtain a predicted yield difference value, and the yield of the to-be-measured year is obtained by combining the yield of the reference year. The application realizes explicit depiction of interannual differences, effectively expands training samples, reduces extreme value interference, improves prediction stability and precision, adapts to various deep learning models, and is easy to popularize in large-scale areas.

The application provides a facial expression recognition method based on multi-module collaborative optimization, has an efficient network structure, and can realize high-precision and strong-robust facial expression recognition in a complex environment. The method comprises the following steps: receiving an RGB face image and performing preprocessing to obtain a preprocessed image; inputting the preprocessed image into a hybrid feature network module to output a first feature map; inputting the first feature map into an efficient local attention mechanism module to output a second feature map; and inputting the second feature map into a classifier module to realize facial expression recognition.

The present application relates to a kind of method for realizing D-Vine Copula soft measurement based on physical information and skewness, comprising the following steps: obtaining real sample set;Determine the monotonic relationship between output variable and input variable;Virtual sample is obtained using Latin hypercube sampling;Determine the kind of binary Copula;Using genetic algorithm to optimize the parameter of binary Copula, calculate monotonicity loss on virtual sample;The Copula density value of each sampling point is calculated by calculating the input variable of test sample and each sampling point, and the weight of each sampling point is calculated;The conditional skewness of test sample is calculated, and according to the threshold set, majority prediction or mean prediction is selected.The method, system, device, processor and computer readable storage medium thereof for realizing D-Vine Copula soft measurement based on physical information and skewness of the present application are used for the nonlinear, non-Gaussian, variable coupling relationship and small sample problem of industrial data, establish D-Vine Copula soft measurement, consider physical loss using GA to optimize binary Copula parameter, consider test sample conditional skewness, improve prediction stability.

The application relates to the field of small sample target detection, in particular to a two-stage small sample target detection method based on an optimized CBAM attention mechanism, and comprises the following steps: training a two-stage target detection network Faster-RCNN by using a base class data set to obtain a base class detection model; freezing parameters of a feature extraction backbone network in the base class detection model; optimizing a CBAM attention mechanism module; placing the optimized CBAM attention module in the feature extraction backbone network to construct a detection network, then inputting a new class small sample data set with a small amount of labeled information to fine-tune parameters of a detection head part of the detection network; and inputting a to-be-detected data set into the detection network to obtain a detection result. Compared with the prior art, the application has the advantages of inhibiting the influence of unimportant spatial information, improving the attention degree of important spatial information, enhancing the sensitivity to different scale features, and having strong generalization ability and robustness and the like.

This invention relates to the fields of image processing and artificial intelligence, and particularly to a few-shot object detection method based on support feature diffusion generation. The method includes steps such as constructing a few-shot object detection base class meta-learning training set and grouping image data within the training set. This invention introduces a support feature diffusion generation module, applying a diffusion model to the generation and optimization of support features. During the training phase, the inverse process of adding noise to features is learned, enabling the model to recover robust support feature representations from noise. During the testing phase, through progressive reverse diffusion, a more stable and diverse set of support features can be generated from a small number of support samples, effectively mitigating the feature instability and overfitting problems caused by few samples.

This invention provides a method and system for underwater acoustic target recognition based on Mel-Cepstral Interpretation (MCI) and Attention Residual Networks (ARDNs), comprising: Step 1: acquiring labeled underwater acoustic target samples, extracting MCI features, first-order difference features, and second-order difference features from the samples to construct a three-dimensional MCI feature vector; Step 2: constructing a target recognition network based on an attention residual module; Step 3: training the target recognition network using the MCI features to obtain an underwater acoustic target recognition model, and using this model to identify the target underwater acoustic object to determine its category. This invention not only enhances the temporal information of the original features but also effectively utilizes various excellent recognition networks and pre-trained models in the field of image recognition, exhibiting strong generalization, stability, and high recognition efficiency. Furthermore, it is applicable to underwater acoustic target recognition with different sampling rates and signal lengths.

This invention provides an automatic SEM image classification method and electronic device based on an improved ConvNeXt network model. The method involves acquiring a target image using a scanning electron microscope (SEM); the target image is a microscopic electron microscope image; the target image is input into an improved ConvNeXt network model to obtain the probability of the target image belonging to each image category; the ConvNeXt network model includes a channel attention mechanism (ECA) module and a global attention mechanism (GAM) module; the image category corresponding to the highest probability is determined as the target image category. Compared to existing technologies, by introducing the channel attention module (ECA) and the global attention mechanism (GAM), the interaction information between different channels can be effectively captured during training, and the network pays more attention to important information and suppresses interference from irrelevant information during learning, thus improving the image classification ability.

This invention discloses a fake image detection method based on artifact decomposition. The method includes acquiring the image to be detected and inputting it into a trained three-branch artifact-aware encoder (scene consistency branch, imaging realism branch, and signal naturalness branch) for feature extraction, resulting in three artifact reflection maps. A trained cross-dimensional gated collaborative fusion module is used to fuse the features of the three artifact reflection maps, generating a unified artifact embedding representation. This unified artifact embedding representation is then input into a trained classification head, and the label classification head of the classification head outputs the true / false prediction probability of the image to be detected. This fake image detection method solves the problems of existing technologies, such as inability to uniformly detect multi-source heterogeneous fake images, poor generalization ability, and susceptibility to overfitting.

This invention discloses a recommendation system and method based on category distribution perception and noise-enhanced contrastive learning, belonging to the field of recommendation system technology, to solve the problems of insufficient utilization of category information and poor stability and robustness of representation learning caused by noise interference in existing recommendation systems. The system includes a category distribution perception module, a graph propagation and view generation module, a joint optimization module, and a recommendation module. The category distribution perception module constructs user category distribution vectors and item category attribute vectors to generate initial representations with category priors. The graph propagation and view generation module generates noise-free and noise-enhanced view representations on the user-item bipartite graph, respectively. The joint optimization module integrates the recommendation main task loss and contrastive learning loss to update the representation parameters. The recommendation module calculates user preference scores based on the optimized representations and outputs recommendation rankings or Top-K recommendation results. This invention can effectively improve the stability and robustness of representation learning while ensuring recommendation accuracy.

This invention discloses an automatic method and system for extracting landslide areas from remote sensing images based on few-sample enhancement and multimodal fusion. The method includes: acquiring and preprocessing multimodal remote sensing data; constructing a foreground-aware few-sample enhancement module, expanding the training samples through geometric transformation, spectral perturbation, and random cropping enhancement strategies; constructing a dual-branch feature extraction network to extract spectral and topographic features respectively; achieving adaptive fusion of spectral and topographic information through a cross-modal attention fusion unit; training a deep learning model using a composite loss function including cross-entropy loss, Dice loss, and boundary constraint loss; and performing topographic constraint post-processing and morphological optimization on the model output to obtain the automatic extraction result of landslide areas. This invention can achieve high-precision landslide identification under limited sample conditions, effectively reducing the false detection rate and improving the model's generalization ability and adaptability to complex scenes.

The invention relates to the technical field of drug performance prediction, in particular to a drug performance prediction method and system based on a neural network, and the method comprises the steps: obtaining an SMILES character string of a drug molecule, converting the SMILES character string into a molecular graph, carrying out the feature coding of atomic nodes and chemical bond edges, and obtaining the features of the molecular graph; adopting a BRICS substructure decomposition method to obtain local subgraph features; a drug performance prediction model is constructed, the model comprises a local feature extraction branch, a global feature extraction branch, a feature fusion unit and an integrated prediction unit, local feature vectors and global feature vectors of molecules are extracted respectively, and prediction results are output after fusion; training the model by adopting a staged training strategy, and performing calibration processing based on the calibration set after training is completed; and inputting to-be-tested drug molecules into the model to obtain a drug performance prediction result. The method effectively captures the local and global structure information of molecules, and improves the accuracy and reliability of drug performance prediction.

The application discloses a transformer partial discharge identification method, device and equipment. First, a three-dimensional digital twin model of the target transformer is constructed, and time synchronization processing is performed on the model to obtain a time compensation parameter. Then, the model and the time compensation parameter are used to perform RTM coarse positioning on the sound pressure signals received by each AE sensor, so that the search area of the partial discharge source is determined. In the determined PD source search area, multiple reference sources are selected. Then, with the aid of the pre-constructed PINN, the reference sources are iteratively positioned and signal analyzed to obtain the position of the PD source and the corresponding discharge waveform characteristics. Finally, based on the extracted waveform characteristics, the discharge category is identified, and the PD source position is associated with the corresponding discharge category and output. The application can break through the small sample dependence, realize high-precision monitoring of the transformer partial discharge, and adapt to the transformer operating environment under different working conditions.

This invention relates to the field of emotional robots, specifically disclosing a voice emotion recognition system for an emotional companion robot. The system includes a voice input port embedded within the robot, a large-scale model processing system, and an emotion recognition output port. The voice input port collects voice data, which is then processed by the large-scale model processing system and output through the emotion recognition output port. The large-scale model processing system comprises a voice preprocessing module, a feature extraction module, a feature fusion module, an emotion classification module, and a result feedback module. This system optimizes the voice preprocessing flow, improves anti-interference capabilities, extracts seven types of multi-dimensional features, and comprehensively characterizes emotional information. It designs a 1D CNN network with "3 convolutional layers + 3 fully connected layers" to enhance global feature integration capabilities. It employs a triple data augmentation strategy and an optimized training strategy to improve model generalization ability and training effect. A result feedback module is added to optimize the recognition results in real time.

The present application relates to the technical field of battery curing detection, and more particularly to a method for detecting the curing degree of a semi-solid battery based on machine learning. The technical solution comprises the following steps: obtaining a multi-modal detection data set of the semi-solid battery to be detected; the multi-modal detection data set at least includes ultrasonic data, battery basic attribute parameters and curing process parameters; inputting the multi-modal detection data set into a pre-trained curing degree evaluation model, and directly outputting the curing degree prediction value and its prediction confidence interval of the semi-solid battery to be detected through model inference. By constructing a multi-modal data fusion and physically guided multi-task learning neural network, the present application realizes the comprehensive improvement of the semi-solid battery curing degree detection in detection accuracy, reliability and efficiency, and enables the model to have strong generalization ability and interpretability.