1612results about How to "Reduce bias" patented technology

The invention discloses a multi-working-condition-oriented equipment fault diagnosis method based on unbalanced learning, and the method comprises the following steps: collecting the operation data of equipment under different working conditions through a plurality of sensors, carrying out the preprocessing, feature extraction and feature selection of the collected original operation data, and obtaining a feature set; processing the feature set by adopting an unbalanced learning technology and a cost-sensitive learning algorithm to obtain a balanced data set; constructing a multi-working-condition fault diagnosis model based on learning, and optimizing model parameters by adopting a method of combining cross validation and grid search according to a balanced data set in a model training process; deploying a multi-working-condition fault diagnosis model obtained through training, carrying out the online diagnosis of equipment faults, and continuously updating and optimizing the model through a continuous learning mechanism. According to the method, the unbalanced learning technology is applied to multi-working-condition fault diagnosis, the accuracy and robustness of diagnosis can be remarkably improved, and real-time monitoring and predictive maintenance of the equipment health state are achieved.

The invention relates to the technical field of data processing, and discloses a quality control method and system based on buoy observation data. The method comprises the following steps: performing multi-dimensional anomaly detection processing on preprocessed standard format data to obtain an anomaly detection report marked with abnormal and suspicious data points; performing multi-source data fusion and cross validation processing on the data in the anomaly detection report to obtain an initial data quality score; performing intelligent correction processing on the anomaly detection report to obtain a corrected data set and uncertainty estimation data corresponding to the corrected data set; performing consistency analysis processing on the corrected data set in a space-time dimension to obtain a space-time consistency score and abnormal region marking data; and carrying out comprehensive evaluation processing on the abnormal region mark data to obtain a target data quality grade identifier and a target quality report, and carrying out quality control strategy analysis to obtain a target quality control strategy. According to the invention, the accuracy of quality control based on buoy observation data is improved.

The invention provides an intelligent post matching and employee growth path planning system and method and a storage medium. The system comprises a data acquisition and processing module, an employee portrait construction module, an intelligent post matching module and an occupational path planning module. The data acquisition and processing module collects employee performance data, education background, work experience and other information from internal and external data sources. The employee portrait construction module constructs a comprehensive portrait of each employee by using a clustering analysis and factor analysis technology; the intelligent post matching module performs intelligent matching by applying a recommendation system algorithm according to post demands and employee portraits, and finds the posts most suitable for the current capabilities and development desires of the employees; the occupational path planning module analyzes the industry development trend and future skill requirements based on the prediction model, and provides personalized occupational development paths and training suggestions for employees. According to the invention, historical performance evaluation and industry trend analysis of employees can be integrated, and comprehensive post matching and occupational planning solutions are provided.

The invention provides an aviation video stream target identification processing method and system, relates to the technical field of image processing, artificial intelligence and aviation monitoring, and aims to realize the target identification processing of an aviation video stream through preprocessing modal data such as video, infrared and radar and combining an adaptive weighted fusion technology, a multi-stage visual feature extraction and a global convolution optimization technology. The system can extract target features from multi-level and multi-modal data; and a historical motion track and future motion prediction are further combined, the system dynamically corrects the target position, and finally the consistency of multi-modal data is ensured through cross validation. According to the method, the recognition precision and robustness of the system to weak and small targets are enhanced, missing detection and false detection are reduced, the adaptability is high, the requirement for computing resources is low, and the method is suitable for target recognition and tracking tasks in a complex dynamic environment.

The invention relates to the technical field of wire harness intelligent evaluation, in particular to a wire harness connection performance intelligent test method and system. The method comprises the following steps: collecting wire harness test data generated by a wire harness test device and a wire harness sensor in real time through an edge computing node; performing electrical performance detection extraction and mechanical performance detection extraction according to the wire harness test data to respectively obtain wire harness electrical performance detection data and wire harness mechanical performance detection data; performing electrical performance deep evaluation according to the wire harness electrical performance detection data to obtain wire harness electrical performance evaluation data, and performing mechanical performance deep evaluation on the wire harness mechanical performance detection data to obtain wire harness mechanical performance evaluation data; and performing environment pressure cross reference according to the wire harness electrical performance evaluation data and the wire harness mechanical performance evaluation data to obtain wire harness connection performance test evaluation data. According to the method, the accuracy of wire harness performance evaluation is improved, and reliable performance data can be obtained in different application fields.

The invention relates to the technical field of gas energy monitoring, and discloses a gas energy monitoring method based on Internet of Things, comprising the following steps: S1, multi-source data acquisition: using an intelligent sensor to acquire gas related data, and sending the data to an edge device or a cloud platform through a wireless network; s2, data preprocessing: carrying out data cleaning, denoising and feature extraction on edge equipment, executing multi-source data fusion, carrying out anomaly detection and processing, and filtering abnormal data; and S3, data transmission: performing data transmission by adopting an encrypted communication protocol, and using a data buffering and retransmission mechanism in the transmission process, the data buffering and retransmission mechanism being used for ensuring data integrity. By using a multi-source data fusion technology and an advanced data preprocessing algorithm, sensor noise and data errors are effectively reduced, the quality and consistency of gas data are remarkably improved, and the problem that the analysis accuracy is reduced due to data source diversity and complexity is solved.

The invention discloses an automatic fine ore sorting method and system based on XRT (X ray tomography) and hyperspectrum, which are used for realizing comprehensive fusion of ore surface and internal information by combining X-ray tomography and hyperspectral imaging technologies so as to improve the recognition precision and the sorting efficiency. XRT is used for extracting the internal structure and density characteristics of the ore, hyperspectral imaging is used for identifying the mineral composition characteristics of the ore surface, and information complementation is achieved. The system adopts a multi-modal convolutional neural network to perform independent feature extraction on XRT and hyperspectral data, and the extracted features are fused through a deep residual network. And the fused features are processed by a cross-modal attention mechanism to ensure that key features of different modals are fully integrated and optimized, and finally high-precision ore classification and sorting are realized. According to the method, the limitation of a single data source on feature extraction is overcome, the ore recognition and sorting precision is greatly improved, and the refined and intelligent development of ore sorting is promoted.

The invention discloses an air pollutant concentration prediction method and related equipment thereof, and relates to the technical field of data processing. The method comprises the following steps: acquiring current pollutant concentration data, current meteorological data and target site geographic position data corresponding to a target environment monitoring site; taking the current pollutant concentration data, the current meteorological data and the geographic position data of the target station as target input data of the prediction model; wherein an input module of the prediction model is used for performing modal decomposition processing on input data to obtain multiple pieces of modal data; the time feature extraction module is used for extracting time feature data; the spatial feature extraction module is used for extracting spatial feature data; the space-time fusion convolutional network module is used for extracting space-time features; the output module is used for outputting a prediction result; determining a prediction result of the air pollutant concentration corresponding to the target environment monitoring station according to the output result; the method achieves the effects of improving the comprehensiveness and accuracy of a prediction result and expanding application scenes.

The invention belongs to the field of fault diagnosis, and particularly relates to a main shaft bending common characteristic diagnosis system based on a stacking integration method, and the system comprises the following modules: a sensor module which is used for collecting a vibration signal of a main shaft in real time; the data acquisition and processing module is used for converting an analog signal into a digital signal and preprocessing the signal acquired by the sensor; the feature extraction module is used for extracting features from the preprocessed vibration signals; the machine learning diagnosis module analyzes and classifies the extracted features by using stack integration based on a plurality of trained machine learning models, and judges the health state of the main shaft; and the output module is used for outputting the diagnosis result and sending out an alarm signal when abnormality is detected. Real-time monitoring and automatic diagnosis of the main shaft bending fault are achieved, compared with a traditional method depending on expert experience and manual detection, the fault detection efficiency and accuracy can be greatly improved, and misjudgment and missed judgment risks caused by human factors are reduced.

The invention discloses a contract evaluation method, system and device based on a large model and a medium, and the method comprises the steps: obtaining a target contract, carrying out the data preprocessing, obtaining the preprocessed contract data, and inputting the preprocessed contract data into a preset contract large language model; the preset contract large language model stores a vector knowledge base including coastal bulk cargo transportation knowledge, legal clauses and industry regulations; in a preset contract big language model, performing contract classification and contract summary extraction on the current preprocessed contract data by using a first model agent to obtain contract summary information; performing contract risk assessment on the current preprocessed contract data to obtain risk assessment information and / or risk avoidance suggestion information; performing benefit measurement and calculation on the current preprocessed contract by using a third model agent to obtain benefit measurement and calculation information; according to the method, the examination and evaluation of the large model in the coastal bulk cargo transportation contract are realized, and the purposes of saving the labor cost and improving the business turnover rate are achieved.

The invention relates to the technical field of genomics, in particular to a gene detection data cleaning method and system based on artificial intelligence. The method comprises the following steps: acquiring gene detection data, and performing preliminary filtering on the gene detection data based on removal of low-quality reads to obtain preliminary filtering data; performing linker sequence identification and removal on the preliminarily filtered data, and performing sequencing error correction to generate gene sequence data; performing GC content analysis and repetitive sequence identification on the gene sequence data to obtain sequence feature data; performing dynamic quality threshold adjustment according to the sequence feature data, and performing secondary filtering on the gene sequence data to generate high-quality sequence data; and performing sequence pattern recognition on the high-quality sequence data to obtain sequence pattern feature data. According to the method, deep analysis is carried out by utilizing sequence characteristics such as GC content and nucleotide distribution, and the recognition capability on sequence complexity and variation modes is improved.

The invention discloses a power load model parameter identification method based on improved deep reinforcement learning, and belongs to the technical field of power load identification. The method comprises the following steps: determining a to-be-identified power load parameter based on a complete comprehensive load model considering power distribution network impedance and reactive compensation; then designing a power load parameter identification model based on a bidirectional gating cycle unit depth deterministic strategy gradient algorithm of priority experience playback; circuit data are collected for model training, and the trained model is used for parameter identification. On the basis of an existing power system mechanism model and a data driving model, the load model parameters are identified by improving the deep reinforcement learning model, and verification shows that the method can accurately identify the power system load model parameters, can improve the fitting precision of the power system load model, and has a good application prospect. The operation strategy of the power grid is optimized, and the operation cost is reduced.

The invention discloses an establishment method of a rock multi-field coupling model and the model thereof, and belongs to the technical field of engineering geological survey. Through comprehensive application of multi-scale modeling, a nondestructive testing technology, a machine learning algorithm and intelligent optimization, the problems that existing multi-physical field coupling analysis is not deep enough, the precision of an observation method is poor, and a device is prone to being damaged are solved, the advantages of various observations are combined, the building and analysis precision of a rock multi-field coupling model is improved, and the rock multi-field coupling analysis precision is improved. The method has the advantages that the method is simple and convenient to operate, the intelligence and automation level of the model is enhanced, microstructure data and macroscopic response are combined, deeper understanding of rock behaviors is provided, meanwhile, more accurate and efficient decision support can be provided for rock engineering design and construction, and the method is worthy of popularization and application. And a novel analysis and simulation method is provided for the field of rock mechanics research. Through the comprehensive scheme, behaviors of rocks under complex geological conditions can be better predicted and managed, so that the safety and economical efficiency of engineering are improved.

The embodiment of the invention provides a cylindrical battery, a battery pack and a power utilization device, the cylindrical battery comprises an anti-explosion sheet and a pore plate, the pore plate comprises a main body area and a thinning area, the thinning area is connected with the anti-explosion sheet, the thickness of the main body area is L1, the thickness of the thinning area is L2, 0.2 < = L2 / L1 < = 0.4, the anti-explosion sheet is configured in a manner that the air pressure in the cylindrical battery reaches a first threshold value, the anti-explosion sheet turns over, and the air pressure in the cylindrical battery reaches a second threshold value. And the electric connection with the pore plate is disconnected. Notching grooves do not need to be formed in the thinning area of the pore plate, power-off pressure deviation can be reduced, and power-off pressure stability is improved.

The invention provides a wheat drought disaster remote sensing monitoring and evaluation method in a flowering period, and relates to the technical field of agricultural disaster early warning. Remote sensing image data is collected based on an unmanned aerial vehicle, and a normalized vegetation index, an enhanced vegetation index and a soil moisture index are generated by extracting multispectral reflectivity in the wheat flowering period; generating a comprehensive remote sensing index, establishing a time sequence set, setting time windows in the time sequence set, calculating a change rate of each window, setting a threshold value, identifying a potential drought region, and generating a drought influence index by analyzing drought duration and an image range; finally, a drought influence threshold value is generated in combination with historical data to judge the drought condition of the current monitoring area.

The invention belongs to the technical field of Internet of Vehicles and communication security, and relates to a vehicle intrusion detection system and method based on a Hybrid EfficentNet model, and the system is obtained by training the Hybrid EfficentNet model. Wherein the Hybrid OfficientNet model comprises an initial convolutional layer, a backbone network, a self-adaptive global pooling layer, a Flaten layer, a discarding layer, a hidden layer and a classification layer; the backbone network is composed of one Fusion MBConv module and 15 MBConv modules, and each MBConv module comprises an expansion layer, a depth separable convolution module, a channel and space attention mechanism and a projection layer; residual connection is used in the MBConv module; the system has excellent capability of detecting abnormal traffic, realizes high accuracy, high-efficiency processing capability and high response speed, meets the requirement of convenient operation in a vehicle environment, and effectively solves the technical problems that an existing intrusion detection model is complex and inconvenient and cannot be directly deployed in a resource-limited vehicle-mounted network environment.

The invention provides a micro robot motion control method and system, and the method comprises the steps: carrying out the synchronization and calibration processing of environment state information, and generating a comprehensive perception image; analyzing the comprehensive perception image to obtain a classification result, and predicting the environment state change influencing the motion track of the micro robot in a period of time in the future based on the classification result; generating a candidate path by using a fast random tree, and optimizing the candidate path in combination with an AI search algorithm to obtain a target path; constructing a three-dimensional dynamic probability graph, continuously updating the three-dimensional dynamic probability graph by using a reinforcement learning algorithm, and generating an instant decision map; and estimating the position of the micro-robot by using a Kalman filter to obtain the position state of the micro-robot, adjusting the speed and steering angle of a driving motor of the micro-robot, and generating trajectory tracking of the micro-robot. According to the technical scheme provided by the invention, the flexibility and intelligence of the micro robot are improved.

The invention belongs to the technical field of welding seam detection, and particularly relates to a laser welding seam defect detection machine which comprises support plates, adjusting assemblies arranged between every two adjacent support plates, driving assemblies installed on the support plates and balancing assemblies arranged on one sides of the support plates. Wherein three groups of support plates are arranged in a circumferential array, a front-end support is fixedly mounted on one side of the long edge of each group of support plates, an ultrasonic detection probe, a dot matrix laser and a camera are sequentially mounted on the three groups of front-end supports, an electric control module is mounted on the top surface of each support plate, and a power supply module, a data analysis processing module and a control system are arranged in each electric control module. The device can automatically move on a reducing pipeline, perform positioning detection on a welding seam, and perform circumferential uninterrupted detection in the detection process, so that the detection precision is higher, a mode of combining laser detection and ultrasonic detection is utilized, displacement deviation is automatically corrected after reducing, the error probability is greatly reduced, unnecessary workload is avoided, and the detection efficiency is improved. The detection efficiency is improved.

The embodiment of the invention provides an X-band dual-polarization radar attenuation correction method based on artificial intelligence. The method is applied to the field of X-band dual-polarization radar attenuation correction. The method comprises the steps of performing space-time matching on existing observation data of an X-band dual-polarization radar and existing observation data of an S / C-band radar in each region to generate a space-time matching data set of each region; according to the space-time matching data set of each region, training an attenuation correction model of the corresponding region, and generating an attenuation correction model of each region; and performing quality control on the observation data of the X-band dual-polarization radar to be subjected to attenuation correction, and inputting the observation data into the attenuation correction model of the corresponding region to obtain the horizontal reflectivity and the differential reflectivity of the X-band dual-polarization radar after attenuation correction. Therefore, the signal attenuation of the horizontal reflectivity and the differential reflectivity of the X-band dual-polarization radar can be effectively corrected, the deviation between the X-band dual-polarization radar and the observation value of the adjacent S-band radar can be reduced, the hardware resource overhead can be reduced, and the method can be efficiently applied nationwide.

The invention relates to a pollutant concentration diffusion simulation method based on grid optimization and self-regulation. The method comprises the following steps: carrying out basic grid division on a simulation area to obtain a basic grid; calculating a pollutant concentration field of the simulation area according to the spatial position and the pollutant concentration value of the monitoring station; generating a concentration distribution matrix according to the pollutant concentration field of the simulation area, and then dynamically adjusting the grid density of the basic grid; calculating the calculation complexity of each grid region by using the grid density of the adjusted basic grid, carrying out pollutant concentration calculation task allocation according to a cloud platform and a weighted polling algorithm, and forming a pollutant concentration distribution prediction result of the simulation region by a pollutant concentration prediction result of each grid; and dynamically adjusting a pollutant concentration distribution prediction result according to the Kalman filtering and the obtained real-time environment data to obtain a final pollutant concentration prediction value. By adopting the method, the precision and the real-time performance of pollution diffusion simulation can be improved.

The invention discloses a method for automatically generating a medical report by using multi-modal feature fusion Mamba. The method comprises the following steps: inputting an original medical image into a visual feature extraction module to output visual features; inputting the original medical report into a text feature extraction module, and outputting text features; respectively inputting the visual features and the text features into an SSM-Attention visual text feature integration module, and outputting new visual features and text features; inputting the new visual features into a Transform encoder, enhancing the visual features, and extracting advanced visual features; and inputting the text features and the enhanced advanced visual features into a multi-modal fusion Mama module to output fusion features, and inputting the fusion features into a Transform decoder to automatically generate a medical report. According to the invention, the SSM-Attention and the Multi-modal Mama can be independently developed and improved, newly developed technical modules can be conveniently integrated, and the method is not only limited to fusion of medical image and text features, but also suitable for fusion of any multi-modal features.

The invention provides a digital twin model correction method and device and a computer program product. The method comprises the steps that an initial digital twinborn model is obtained, and the initial digital twinborn model is a virtual model of a fully mechanized coal mining face constructed based on the digital twinborn technology; a correction coefficient is obtained, and the correction coefficient is obtained according to the change of the coal mine fully mechanized coal mining face after the initial digital twinborn model completes the preset instruction; and correcting the initial digital twinborn model according to the correction coefficient to obtain a target digital twinborn model. According to the scheme, the problem that in the prior art, the state after the twin model responds to the completion action instruction is deviated from the state of the real coal mining environment is solved.

The invention discloses a satellite-ground collaborative convection inception identification and forecasting method based on physical identification, and belongs to the technical field of convection inception identification and forecasting based on artificial intelligence. The prediction performance of the artificial intelligence prediction model can be effectively improved through more accurate prediction input samples, and the empty report rate and the false report rate of the prediction model are reduced. According to the method, based on multi-source observation and forecast data of satellites, ground-based radars and the like, high space-time matching is carried out on a satellite-based CI objective judgment result, a radar CI recognition result and ground minute-level rainfall data, and a more accurate high-quality convection inception annotation data set is formed; on the basis, a CI environment parameter matrix based on physical judgment is obtained by combining a numerical model product and utilizing a deep learning neural network model, and is fused with satellite parameters, model precipitation and the like based on a CI annotation data set, so that satellite-ground collaborative high-quality forecasting factors are provided for CI intelligent forecasting model training, and the empty report rate and the false alarm rate of CI are effectively reduced.

The invention discloses a construction method, system and equipment of a power dispatching knowledge question-answering system and a storage medium. The construction method comprises the steps of obtaining a power dispatching semantic document needing to be processed by the power dispatching knowledge question-answering system; sequentially carrying out character-based segmentation, proposition conversion and proposition combination on the power dispatching semantic document on the basis of the large language model and the related cue words to obtain a plurality of text blocks; based on the large language model and the related prompt words, identifying and extracting entities related to power dispatching and relationships among the entities from each text block; constructing a local graph for each text block on the basis of the large language model and the related prompt words, the entities extracted from each text block and the relationship among the entities; traversing each text block, and combining all the local graphs into a global graph; and on the basis of the large language model and the related prompt words, according to the question type queried by the user, a reply text is correspondingly generated by matching a minimum local graph which is most related to the global graph in the global graph.

The invention relates to a wind power cabin equipment quality evaluation and maintenance decision-making method based on reinforcement learning, and belongs to the field of wind turbine generator operation and maintenance. According to the method, the state degradation model of the wind power cabin is established by using the Markov decision process, and the state transition probability is estimated in combination with the dynamic integrated probabilistic neural network, so that the prediction accuracy is improved. Weibull distribution parameter estimation is carried out on the fault data of the subcomponents of the wind turbine cabin through the grey model, life reliability analysis of the components is completed, then a reliability function of the wind turbine cabin is obtained, and equipment quality evaluation is achieved. According to the method, a maintenance method combining post maintenance and opportunity maintenance is adopted, an objective function of a maintenance decision model is established according to a component reliability function and a maintenance cost model, and an optimal maintenance strategy is solved by using an AD3QN algorithm, so that the minimum maintenance cost is realized, and the operation efficiency and economic benefits of the wind power plant are improved.

The invention relates to a method for identifying and extracting a potential sliding surface of a side slope under three-dimensional numerical simulation analysis, and belongs to the field of traffic side slope engineering. According to the method, a strength reduction technology does not need to be introduced, a rock-soil body shear strength criterion is not limited, after the slope stress field is obtained through numerical simulation calculation, the three-dimensional slope potential sliding surface can be efficiently, automatically and intelligently recognized and extracted, the obtained potential sliding surface conforms to a slope shear failure mechanism, and the method is suitable for large-scale popularization and application. The actual damage form of the three-dimensional slope in a real scene can be effectively displayed; according to the method, the sliding circumference boundary line of the three-dimensional potential sliding surface is determined firstly, and then the discrete points of the three-dimensional potential sliding surface are extracted within the range, so that intelligent identification and extraction of the three-dimensional potential sliding surface are facilitated. The method has the advantages of being wide in application range, high in efficiency, small in deviation, high in universality, good in effect, low in cost and the like, has the automatic and intelligent characteristics, and can provide a powerful scientific basis for slope potential instability range prediction and reinforcement area determination in actual engineering.

The invention provides a three-phase current sampling circuit calibration method and system, and belongs to the technical field of permanent magnet synchronous motor three-phase current sampling calibration, and the method comprises the following steps: sequentially giving UVW three-phase current through a motor controller; respectively acquiring three-phase current data through a current sampling circuit in the motor controller and an external current sensor; performing linear fitting on the acquired three-phase current data to obtain a current sampling compensation coefficient of the UVW three-phase current sampling circuit; and multiplying the current sampling compensation coefficient by an initial sampling coefficient of the current sampling circuit, and calculating to obtain an actual sampling coefficient. The current sampling compensation coefficient can more accurately reflect the performance of the current sampling circuit under the actual working condition, so that errors and deviation are reduced, and the sampling precision of the current sampling circuit under the large current is improved.

The invention provides a robot collaborative sewing method and system based on speed look-ahead control. The system comprises an image detection module, a pressure detection module, sewing equipment, a controller and a mechanical arm. The image detection module acquires image information of the fabric in real time and transmits the image information to the controller; the controller processes the image information to generate the position of a pressing point at the tail end and a motion path, introduces speed look-ahead control, optimizes a speed extreme value according to the motion path, completes speed planning of the whole track, and interpolates the planned track to obtain an interpolated result; the controller sends expected force to the pressure detection module, and the pressure detection module outputs the pressing height; the mechanical arm presses the height to execute the action to the fabric based on the interpolation result; the controller sends a control signal to the sewing equipment, and the sewing equipment carries out cloth feeding and sewing; the speed of the sewing equipment is matched with that of the mechanical arm, and cooperative sewing is carried out.

The invention relates to the technical field of gene analysis bioinformatics, and particularly discloses a gene data analysis method based on a large language model.The method comprises the steps that the quality of gene data to be analyzed is analyzed, whether the gene data to be analyzed is available or not is judged, and a gene sequence segmentation scheme of the available gene data is determined; inputting the gene data after gene sequence segmentation into the large language model to obtain a gene data association feature data set and a gene data analysis result, determining a gene data analysis accuracy feature value, and judging whether the gene data analysis result conforms to expectation or not; the problems that quality and integrity problems of traditional biological data challenge the accuracy of analysis results and effective integration and analysis of multi-modal data still need to be improved are solved, the data processing efficiency is optimized in the whole process, repeated calculation, misjudgment and invalid results are reduced, the analysis efficiency is improved, and resource waste is reduced.

The invention discloses a method for detecting internal body defects of a crystal. The method comprises the following steps: S1, taking the detected crystal with the internal body defects; s2, irradiating from the side surface of the crystal by using a line light source so as to obtain a crystal tomographic image containing scattered light of body defects in the crystal; correcting the size of the crystal chromatography image; s3, performing time sequence analysis and processing on the corrected crystal chromatography image, and highlighting the characteristics of defect points; s4, in combination with the internal defect position of the known crystal and the detection time sequence signal of the position, training the RNN model to obtain a crystal internal defect detection model; s5, repeatedly training to obtain a final crystal internal defect detection model; and S6, inputting the time sequence signal of the crystal to be detected, which is obtained by processing in the steps S2 to S3, into the final model, so as to obtain the position information of the internal defect. And S7, filtering out false defects according to the representation of the defects on the plane. The method solves the problems of complex identification process and low identification efficiency of the internal defect of the crystal at present.