194 results about "Real time classification" patented technology

The invention discloses an intelligent monitoring system for a municipal drainage pipe network, and particularly relates to the technical field of drainage pipe network monitoring. The node operation mode identification module carries out real-time classification and confidence evaluation on the operation state of the pipe network, constructs a multi-attribute pipe network weighted graph based on pipe diameter difference, gradient and confluence density, and extracts multi-scale features through graph Fourier transform. A hybrid anomaly detection link is constructed in combination with an LSTM self-encoder, an isolated forest model and chemical oxygen demand and turbidity water quality verification, and the problems that traditional single-index monitoring is prone to false alarm and missing alarm and inaccurate in positioning are solved; sensor data compensation is realized through cooperation with digital twinning, a rapid detection mode is started during rainstorm early warning, key nodes are processed preferentially, and drainage scheduling is controlled in a closed-loop mode; and target nodes which are easy to accumulate grease are screened based on pipe network topology connectivity, accumulation risks are predicted by fusing multi-sensor data, and a preventive clearing instruction is triggered.

The invention discloses an intelligent shield muck improvement method and system based on machine learning, and the method and system achieve the intellectualization and precision of muck improvement in the tunneling process of a shield tunneling machine through integrating muck image classification, improvement parameter prediction and model optimization. Firstly, a YOLOV8 network is utilized to classify muck images in real time, classification randomness is corrected in combination with an XGBOOST model, and classification accuracy and stability are improved. Secondly, performing time sequence prediction on the muck improvement parameters based on an LSTM network, and introducing a linear regression and genetic algorithm to improve the prediction precision; and finally, optimal solution is carried out on the muck improvement parameters through a particle swarm optimization (PSO) algorithm, and the stability and efficiency of shield tunneling are remarkably improved. The method solves the problems of randomness and hysteresis caused by dependence on manual adjustment in the prior art, has the advantages of high efficiency and accuracy, and is suitable for automatic application scenes of shield engineering.

The invention discloses an unmanned aerial vehicle real-time vegetation classification system and method based on a lightweight AI model, and belongs to the technical field of vegetation monitoring and analysis. On the basis of an existing processor of the unmanned aerial vehicle, vegetation real-time classification is achieved by optimizing the deep learning model, and the flight path is dynamically adjusted according to the classification result. The method comprises the steps of system configuration and initialization, lightweight AI model development and optimization, real-time data acquisition and processing, adaptive flight path adjustment and feedback optimization. The method has the following advantages: (1) real-time classification of vegetation and adaptive flight path adjustment are realized, and the acquisition efficiency is optimized; (2) the sampling density is adaptively adjusted for different value regions, and the precision and coverage efficiency of vegetation classification are improved; (3) the acquisition strategy is optimized in real time in combination with the battery state, and unnecessary flight time and energy consumption are reduced; and (4) the method is suitable for application fields such as agricultural monitoring, forestry management and ecological protection.

The invention discloses a multi-modal learning automatic complaint content analysis and classification method and system, and relates to the technical field of semantic analysis, and the method comprises the steps: collecting multi-modal complaint data, and carrying out the preprocessing of the collected multi-modal complaint data; extracting features of the preprocessed multi-modal complaint data, and fusing the features of different modal data; constructing a complaint content classification model by using the fused features, carrying out model training, and evaluating and optimizing the trained model; and deploying the optimized model for real-time classification. According to the multi-modal learning automatic complaint content analysis and classification method provided by the invention, the complaint content is quickly and accurately classified, and a targeted processing strategy is provided according to the main category, the emotional tendency and the specific theme of the complaint. And an emergency response mechanism is automatically triggered to ensure that problems are processed in real time and effective utilization of resources is ensured. Decision rules are continuously optimized from historical data learning, and the intelligent level of a processing strategy and the customer satisfaction degree are improved.

The invention relates to the technical field of intelligent data management, in particular to a multi-source data fusion type intelligent data management system based on edge computing. The system comprises a multi-modal data acquisition unit which acquires multi-source heterogeneous data in real time; the edge preprocessing and enhancement unit integrates a long short-term memory network and a cross attention mechanism, and the lightweight anomaly detection unit deploys a structured pruning and quantized double-branch hybrid model at edge nodes, and performs real-time classification and anomaly scoring on multi-modal fusion feature tensors; and the grading early warning unit is used for receiving the fused comprehensive abnormal score, mapping the comprehensive abnormal score into a discrete alarm grade, and generating a standardized message body containing the grade, a timestamp and details. According to the design, through local tolerance self-adaption and LSH Hash accelerated matching, the coupling complexity of the electrocardio-blood pressure-blood glucose multiple signals is quantified, and the problem that the complex coupling relation among multiple physiological parameters is difficult to capture through a single signal is solved.

The invention discloses an integrated learning classification method and system with a concept drift detection function, and relates to the technical field of machine learning and data stream processing, and the method comprises the steps: obtaining high-dimensional data of an extensible sliding window, carrying out the dimension reduction of the high-dimensional data of the extensible sliding window through PCA, and carrying out the extraction to obtain key feature data; performing real-time monitoring on the key feature data based on an adaptive window early warning detection method, automatically expanding an adaptive window during drifting early warning, and obtaining original data of the adaptive window when concept drifting occurs; and inputting the original data of the adaptive window when the concept drift occurs into a pre-established learning model based on an integrated Hough tree to obtain a trained learning model based on the integrated Hough tree, and performing real-time classification of data streams based on the trained learning model based on the integrated Hough tree. The accuracy of stream data classification can be improved, and the adaptive ability of the model to concept drift is enhanced.

The invention discloses a writing brush calligraphy practice correction system based on real-time handwriting analysis. The system comprises a sensing layer used for collecting handwriting tracks, physiological signals, environmental parameters and ink mark characteristics; the edge calculation layer is used for executing noise filtering, coordinate system normalization, multi-modal data alignment and feature primary extraction through an Apache Kafka data pipeline; and the algorithm analysis layer is used for carrying out super-long calligraphy stroke sequence modeling by adopting an S4 architecture time sequence model, introducing a Neural ODE module for modeling, capturing dynamic characteristics in a continuous pen wielding process, constructing a multi-physics-field coupled PINN framework, constraining neural network prediction through a physical loss function, deploying a dual-stage characteristic extractor to extract high-order characteristics, and extracting the high-order characteristics. Book style features are extracted, and the current practicing book of the user is classified in real time; the intelligent correction layer is used for generating a dynamic correction suggestion, constructing a personalized learning path, realizing self-adaptive scoring and providing aesthetic dimension feedback at the same time; and the user interaction layer is used for providing an AR correction interface.

A system for real-time detection and mitigation of morphing malware in high-density edge networks, consisting of: a data acquisition unit configured to receive, normalize, and encode multimodal telemetry data streams originating from at least one of the following domains: network traffic, process behavior, system call sequences, binary instruction traces, and control flow graphs; the data acquisition unit is further configured to compute feature embeddings over sliding time windows and apply privacy-preserving redactions prior to storage; a generative neural processor that is operationally coupled to the data acquisition unit and configured to generate synthetic morphing malware variants by learning probabilistic transformations of previously observed malicious data representations, maintaining semantic functionality while varying structural and behavioral features; a discriminative neural processor trained adversarially with the generative neural processor, wherein the discriminative neural processor is configured to detect morphing malware by evaluating a probability distribution over multimodal telemetry embeddings and classifying anomalous process and flow behaviors in real time; a coordination processor that is communicatively connected to both the generative neural processor and the discriminative neural processor and is configured to orchestrate adversarial co-training, regulate detection thresholds, calculate reinforcement-based penalties for false negative results, and trigger countermeasures as soon as a detection confidence level exceeds a predefined adaptive threshold; a secure, system-integrated inference and enforcement unit configured to perform low-latency countermeasures at the network edge, including selective packet filtering, flow isolation, process interruption, or system microsegmentation, based on instructions from the coordinating processor; and a hardware-embedded security enclave that is embedded in the system and configured to store cryptographic keys, neural model parameters, and integrity affirmation data to ensure the confidentiality, authenticity, and immutability of model artifacts and policy configurations.

The invention provides a radar track classification method based on a double-layer LSTM model. The radar track classification method is used for classifying radar targets such as unmanned ships, unmanned aerial vehicles and birds in real time. According to the method, in combination with kinematics characteristics of the variable-length radar track, a self-adaptive data preprocessing strategy is adopted, forward filling and backward truncation are carried out on the track, and the effectiveness of characteristic data is ensured through a Masking technology and standardized normalization processing. A double-layer long-short-term memory (LSTM) model is used for time sequence modeling, a mixed attention mechanism is introduced to improve the extraction capability of the model on key track information, abnormal points and non-key track points in data are effectively processed, a batch processing strategy is adopted to improve the real-time performance of the system, and the method is suitable for monitoring unmanned ships and other low-altitude targets.

The invention discloses an online AOI detection system based on an industrial intelligent sensor, and the system comprises an image collection and preprocessing module which is used for collecting and preprocessing image data of an industrial product; the mask auto-encoder modeling module is used for constructing a mask auto-encoder model; the structure parameter optimization module is used for optimizing the mask auto-encoder model; the image reconstruction and difference extraction module is used for generating a reconstructed image, extracting an image difference region and determining a candidate defect region; the defect identification module is used for carrying out edge extraction and aggregation analysis, identifying a final defect area and acquiring spatial position information; the defect classification and labeling module is used for extracting defect area features and generating corresponding classification labels and grade labels; and the control response module is used for generating a control instruction and issuing the control instruction to the production line control device. According to the method, the high-precision automatic identification and real-time classification processing of the surface defects of the industrial product are realized by fusing the mask auto-encoder and the Tiancattle herd optimization algorithm.

The invention provides a game accelerator packet loss optimization method and device based on flow priority management. The method comprises the following steps: executing periodic link detection on at least one proxy node, and generating a corresponding link monitoring result; judging whether the packet loss rates of two continuous monitoring periods exceed a first threshold value or not according to the link monitoring result; performing real-time classification on the to-be-sent data streams to obtain priority labels for different traffic categories; inputting a prediction algorithm based on a link monitoring result, obtaining a network quality prediction value of a next monitoring period, adaptively adjusting a redundancy proportion according to the network quality prediction value, and sending a queue bandwidth constraint parameter and an agent node; and sending the sorted source data packet and the redundant data packet to a target game server or a game client according to the updated proxy node so as to realize accelerated forwarding of the game data. According to the invention, the reliability, delay stability and bandwidth utilization rate of game data transmission in a cross-border high packet loss network environment can be improved.

The invention discloses a sleep real-time staging modeling method, a sleep real-time encoding and decoding method and a sleep real-time staging modeling system. EEG, EOG and EMG signals of a subject are collected, statistics, frequency domain and frequency domain characteristics and other characteristics are extracted after preprocessing, sleep stages marked by experts serve as real labels, a model is trained in a supervised learning mode, and real-time classification of the sleep stages is achieved. Furthermore, high-precision encoding and decoding of sleep content are realized by applying stimulation prompt during sleep, collecting and preprocessing whole-brain EEG signals, distinguishing NREM and REM stages, training encoding and decoding models respectively, and aligning nerve characterization during waking and sleep by utilizing comparative learning.

The invention discloses a rapid identification method applied to defibrillation requirements of an AED device, and belongs to the technical field of medical equipment. According to the method, multiple links including acquisition, preprocessing, feature extraction, mode classification, real-time classification and defibrillation triggering work cooperatively, and advanced signal classification and mode recognition technologies are introduced, so that efficient and accurate electrocardiosignal analysis is realized in the AED device, the response speed and accuracy in the emergency treatment process are remarkably improved, and the emergency treatment efficiency is improved. Multi-dimensional features of electrocardiosignals are extracted and input into a pattern classification model, automatic recognition of arrhythmia is achieved, whether defibrillation operation needs to be carried out or not can be accurately judged, the possibility of misjudgment is reduced, self-optimization can be carried out according to new electrocardiosignal data through a dynamic learning and retraining mechanism, and the accuracy of defibrillation is improved. The adaptive capacity to a complex heart rhythm mode is improved, the first-aid response time is effectively shortened, the safety of a patient is guaranteed, and the first-aid efficiency and the treatment effect are improved.

The invention discloses a numerical control machine tool machining state monitoring method and device, and belongs to the technical field of digital machine tool control. The method comprises the following steps: step 1, constructing a double-layer sensing network system consisting of an environment layer and a machine tool layer, and realizing preliminary physical isolation of environment parameters and a machine tool thermal state; 2, forming a multi-dimensional parameter mapping library of environmental influence and self-heating; 3, decomposing the temperature change signal into a low-frequency environment component and a high-frequency processing thermal response component, respectively extracting environment characteristics and processing characteristics in the low-frequency environment component and the high-frequency processing thermal response component, and realizing real-time classification and decoupling of the temperature signal; 4, constructing a heat influence cross matrix, and quantitatively describing the influence degree and interaction effect of the environmental parameter change and the machine tool heating on the precision of each shaft; and 5, designing a dual-channel parallel compensation control mechanism based on an analysis result of the thermal influence cross matrix, generating a final compensation instruction, and implementing and executing the final compensation instruction to realize precision stable control.

The invention discloses a browser front-end component intelligent classification method and system based on element perception, and relates to the technical field of front-end component intelligent classification, and the method comprises the following steps: executing dynamic structure completion on an incompletely rendered DOM region through a preset structure prediction model based on a trajectory data stream, and generating an enhanced DOM structure chart; extracting first data of each candidate component area in the enhanced DOM structure chart, the first data including structural features, behavior features and visual features, and generating a component semantic embedding vector; constructing a context semantic map based on the component semantic embedding vector, outputting an initial classification result, and performing context correction on the initial classification result; and mapping the corrected initial classification result to a front-end interface, and outputting a real-time classification result. According to the method, a loading track data flow modeling and structure dynamic complementing mechanism based on a browser operation environment is introduced, so that the structure restoration accuracy and modeling robustness of a complex component area in a front-end page are remarkably enhanced.

The invention discloses an automatic warehousing system and a sorting method thereof, and the system comprises an execution module, a sensing module and a digital twin management module: the execution module is composed of a vertical lifting container, an autonomous mobile robot and a sorting machine; the sensing module collects cargo full life cycle data through a visual sensor, an RFID tag and a reader-writer. The digital twinborn management module constructs twinborn bodies of the cargos and the equipment, dynamically classifies the cargos based on real-time data, automatically responds to inventory abnormity and simulates and generates an optimal warehousing and sorting route. According to the sorting method, data are collected through the Internet of Things and the RFID technology, goods are classified in real time through a fuzzy clustering algorithm, the digital twinborn body serves as a training environment, a multi-device collaborative optimal sorting path is generated through a reinforcement learning model, and dynamic adjustment is conducted according to the real-time state. According to the scheme, intelligent management and efficient sorting of the warehousing system are achieved, the inventory management precision and the equipment cooperation efficiency are improved, and the system is suitable for high-frequency and high-flexibility warehousing scenes.

The invention discloses an automatic test input generation method based on dynamic page splitting, and belongs to the technical field of dynamic webpages. In order to effectively verify the semantic consistency of a page in a complex Web application supported by multiple languages, the method comprises the following steps: acquiring a dynamic webpage to be tested, and analyzing a document object model structure of the dynamic webpage based on a Cypress framework; the monitored dynamic content elements are classified and marked in real time by using the real-time DOM operation capability of Cypress; on the basis of the structural information of the page components, in combination with integration of Cypress and GraphQL API, dynamically obtaining test data from a background, and automatically generating a test input template matched with the current state of the page; the method is applied to dynamic page testing in real time; generating a test report; a natural language processing module is introduced, and text content in the page is analyzed; and optimizing a page splitting strategy, an input generation rule and GraphQL query configuration.

Disclosed is a computer-implemented method for real-time classification of cells in a body tissue that includes capturing one or more images of a cytology slide sample of the body tissue, pre-processing the captured one or more images, detecting and extracting a plurality of regions of interest in the pre-processed images using a cell detection process, wherein each region of interest includes one or more cells or cell clusters of interest, segmenting each individual cell or cluster in the regions of interest and extracting pixel content of each segmented cell or cluster, extracting semantic features from each segmented cell image using an embedding extraction model, classifying each individual cell or cell cluster based on corresponding extracted sematic features, and predicting a class of the cytology slide sample, based on one or more positions and relationships among one or more classified cells, using a slide level decision model.

The intelligent strip steel plate shape regulation and control method based on the self-adaptive fusion model comprises the steps that 1, historical rolling production process data are collected and preprocessed; 2, constructing a rolling production process data set based on the preprocessed data; 3, plate shape quality classification is conducted according to the ratio of the strip steel convexity to the target thickness, and category labels are set; 4, establishing a self-adaptive strip steel outlet strip shape diagnosis model containing a plurality of classifiers based on the DS theory, and training the diagnosis model through the rolling production process data set; 5, inputting rolling production process data under a new rolling schedule into a classifier of the trained adaptive strip steel outlet strip shape diagnosis model to obtain a real-time classification prediction result of the strip steel, and if the prediction classification is under-convexity or over-convexity, executing the step 6; otherwise, the process parameters are not adjusted; and 6, according to a prediction result in the step 5, controlling ILQ to dynamically adjust and optimize process parameters based on an inverse linear quadratic form.

The invention provides a current fluctuation anomaly detection method based on machine learning LSTM (Long Short Term Memory) and SVM (Support Vector Machine). The method comprises a data acquisition and preprocessing step, a feature extraction step, a machine learning model training step, an anomaly detection and alarm step and a remote monitoring step. Wherein the machine learning model training step comprises SVM training and LSTM training; the anomaly detection and alarm comprises the following steps: SVM real-time classification: inputting a signal which is acquired and preprocessed in real time into a trained SVM model, and outputting an alarm signal if the SVM model judges that the running state of the motor is abnormal; and LSTM trend prediction: inputting the calculated time sequence data into the trained LSTM model, and if the LSTM model predicts that the future current has abnormal fluctuation, sending out an early warning signal in advance. According to the invention, real-time monitoring and anomaly detection of the running state of the motor can be realized, the safety and stability of equipment are improved, the risk of fault shutdown is reduced, and the production efficiency is improved.

The invention relates to a single-machine obstacle avoidance method and system based on a millimeter wave radar and a camera, and the method comprises the steps: transmitting a frequency modulation continuous wave signal through the millimeter wave radar, collecting and preprocessing environment original data, and outputting a target list; based on the target list, when the target meets a preset signal-to-noise ratio condition and the speed exceeds a preset speed threshold value, triggering a camera to collect an image, performing real-time classification on the image through a lightweight convolutional neural network model, and outputting a classification result and a bounding box; performing space-time alignment on a target list and a classification result, dynamically allocating weights of radar data and visual data by adopting a confidence weighted fusion algorithm, associating the same target through distance calculation, and outputting a fused target attribute; based on the fused target attributes, constructing a three-dimensional decision matrix, and generating an obstacle avoidance strategy according to a preset priority rule; and converting the obstacle avoidance strategy into a bus control instruction, and controlling the vehicle to execute a corresponding alarm or brake response action.

The invention relates to a self-adaptive retrieval method based on multi-modal graph indexing and privacy calculation, and belongs to the technical field of safe and intelligent retrieval. The method comprises the steps that retrieval problems are classified in real time through a complexity discriminator; the method comprises the following steps: disassembling a retrieval problem into multi-modal metadata through atomic memory nodes, and constructing a semantic association map based on a predefined meta-path rule; through a semantic matching module driven by a neural network model or a large language model, multi-type association is established for the newly-added nodes and historical nodes; triggering index field rewriting and topological structure adjustment of old nodes according to newly input retrieval concepts and relationships, and generating an optimal retrieval path; injecting reversible differential privacy noise into the local embedding model; and establishing a homomorphic encryption cache pool for a high-frequency query result. And the structure is flexibly adjusted in the retrieval process, so that the retrieval path better meets the actual demand, the privacy protection strength can be intelligently adjusted according to the content sensitivity, and the retrieval response speed is further accelerated.

The invention discloses a wiring terminal defect detection method, a storage medium and execution equipment. According to the method, the detection precision is remarkably improved by fusing current and image multi-modal data. The method comprises the following steps: firstly, acquiring steady-state current and image data of the wiring terminal in no-load and load stages, constructing a current time sequence curve, decomposing original data, and optimizing image quality by adopting self-adaptive illumination compensation; the core innovation lies in that a dynamic attention fusion network is introduced, dynamic weighted splicing is carried out on current feature vectors and image feature vectors, and real-time classification feedback is realized in combination with a hybrid model. The first feature data set and the second feature data set are subjected to weighted splicing fusion through the dynamic attention fusion network to perform defect prediction judgment and component, and an adaptive illumination compensation algorithm is combined to enhance the image defect edge and suppress reflective interference, so that the data quality is improved from the source, and the defect detection precision is further improved.

The invention belongs to the technical field of action recognition, and particularly relates to a dynamic graph convolution swimming stroke recognition system and method based on mutual information, and the method comprises the steps: carrying out the equally-divided rating of a sequence value interval of an obtained six-axis IMU signal, and obtaining the rated and quantized sequence data; calculating a standardized mutual information matrix of the input IMU signal according to the sequence data, and estimating an adjacent matrix through Laplacian normalization; the adjacent matrix is transmitted to a pre-trained graph convolution model to serve as a calculation unit of graph convolution, and feature representation of the six-axis IMU signals in real-time classified monitoring of different swimming strokes is obtained; the graph convolution model is a dynamic graph convolution neural network MIC-GCN model constructed based on a mutual information matrix; and based on the obtained features, classifying and predicting the output swimming stroke category as an output result label. The method can overcome the defect that the traditional deep learning model is low in swimming stroke category discrimination accuracy in the monitoring process of the six-axis IMU signal.

The present invention discloses an intelligent rehabilitation assistance training system for spinal degenerative diseases based on deep learning. The system of the present invention includes a real-time classification module for traditional Chinese medicine guiding techniques videos based on deep learning and a video sequence division and evaluation module based on human skeleton representation; the former obtains two-dimensional human skeleton data as the training data of the learning model, conducts deep learning training to obtain a generalized deep learning model, and finally obtains the real-time frame classification result; the latter, according to the frame classification result, segments and corrects the skeleton sequences of the same category in real time, and compares and scores the segmented sequence segments with the skeleton sequence segments of the expert group videos of the corresponding category. The system of the present invention does not require the guidance and intervention of medical staff, enables patients to perform traditional Chinese medicine guiding techniques training by themselves at any time, is applicable to families and primary medical and health institutions, can relieve the pressure of medical staff, and improve the flexibility and accuracy of patients' rehabilitation training.

In various examples, the present disclosure relates to using temporal filters for automated real-time classification. The technology described herein improves the performance of a multiclass classifier that may be used to classify a temporal sequence of input signals—such as input signals representative of video frames. A performance improvement may be achieved, at least in part, by applying a temporal filter to an output of the multiclass classifier. For example, the temporal filter may leverage classifications associated with preceding input signals to improve the final classification given to a subsequent signal. In some embodiments, the temporal filter may also use data from a confusion matrix to correct for the probable occurrence of certain types of classification errors. The temporal filter may be a linear filter, a nonlinear filter, an adaptive filter, and / or a statistical filter.

The invention discloses a sparse view angle multi-human body joint reconstruction method based on a dynamic graph convolutional network. The method comprises the following steps: firstly, acquiring indoor RGB video data by using an RGB camera; secondly, based on RGB video data, estimating a human body 3D posture by adopting a human body parameter model estimation algorithm, and performing generation and prediction by utilizing a completion algorithm to complete reconstruction of a human body model; and finally, real-time classification, positioning and tracking of multiple human body actions are realized through an action recognition algorithm, and behavior modes and interaction intentions of family members are analyzed based on the reconstructed human body model and human body 3D postures. According to the method, more flexible human body dynamic representation is realized, the problem of possible inaccuracy of initial posture estimation is solved, and the precision and stability of multi-human body reconstruction are remarkably improved.

The invention discloses a construction facility positioning and classification method and system based on laser radars. The method comprises the steps of arranging the laser radars to obtain construction environment point clouds and sorting the point clouds; taking the point cloud with the maximum serial number as a post-order point cloud, and fusing the post-order point cloud to the pre-order point cloud; taking the fused point cloud as a post-order point cloud, and fusing the post-order point cloud to the pre-order point cloud until the sequence of the fused point cloud is 1; obtaining each sub-point cloud based on a clustering threshold; determining each sub-point cloud coordinate, and constructing a sub-point cloud data set; constructing a facility classification model; training a facility classification model based on the sub-point cloud data set; and a complete point cloud is obtained in real time, sub-point clouds and coordinates of the sub-point clouds are obtained through separation, the sub-point clouds are input into the trained facility classification model to classify the facilities in real time, and categories and coordinates of the facilities are output. The method has the outstanding advantages of accurate positioning, accurate identification, high reliability and the like.

The invention relates to a lightweight wheat seed classification method fusing a mixed attention mechanism, which comprises the following steps: respectively shooting collected wheat and each seed from two angles of an upward ventral ditch and a downward ventral ditch, taking black light-absorbing flannelette as a background during shooting, and obtaining and constructing a wheat grain image set; constructing a lightweight classification model, and inputting the wheat grain image set into the model for training and verification; obtaining a trained and verified lightweight classification model, and classifying the to-be-classified seeds through the trained and verified lightweight classification model; according to the lightweight classification model, a mixed attention mechanism and a stacked inverted residual convolutional layer are combined, so that the model parameter quantity and the calculation quantity are reduced under the condition that the network performance is not reduced; and more local features are learned by using convolution of different scales, so that a relatively small storage space is occupied while relatively high identification precision is maintained, and real-time classification and identification of the wheat grain image on a future low-performance terminal are facilitated.

The invention discloses a multi-mode arc welding real-time quality evaluation system and method, and belongs to the technical field of intelligent welding quality control. The system comprises a distributed sensing array, an edge computing node and a dynamic quality mapping engine, wherein the distributed sensing array realizes multi-modal data synchronous acquisition through a high-frequency electric signal acquisition module, a molten pool thermal imaging module and an acoustic emission sensor array; gPU acceleration parallel processing is adopted by the edge computing nodes, an electric signal variable coefficient CV is extracted through a pulse cluster segmentation algorithm, and the main resonant frequency f0 of a molten pool is obtained by combining FFT frequency domain analysis; a dynamic quality mapping engine constructs a deep learning model containing a material-process knowledge graph, and real-time classification of defects such as pores, incomplete fusion and cracks is achieved. According to the method, the defect recognition accuracy is improved to 92.7%, the evaluation delay is compressed to be within 200 ms, the evaluation stability is high, and the method is suitable for precision welding quality monitoring in the fields of aerospace, new energy automobiles and the like.