365results about How to "Reduce misjudgment" patented technology

The invention relates to the technical field of aluminum alloy detection, and discloses clamping equipment for aluminum alloy detection, which comprises a base, two electric guide rails are symmetrically mounted on the base, an area for placing an aluminum alloy profile is formed between the two electric guide rails, brackets are mounted on the two electric guide rails, a bearing seat is mounted on each bracket, and elastic rolling pieces for pressing the aluminum alloy profile are mounted on the bearing seats; the electric guide rail can drive the support to drive the elastic rolling piece on the bearing base to roll on the surface of the aluminum alloy profile. And a hardness detection piece is mounted in the elastic rolling piece. Rapid identification of surface unevenness is achieved through the elastic rolling part and the displacement sensor, local hardness can be quantitatively detected through the built-in hardness detection part, internal defects caused by insufficient hardness are prevented, dynamic pressing and friction resistance detection of the profile in the full-length range are achieved in cooperation with the adjustable limiting detection part and the pressure sensor, and the detection precision is improved. The comprehensiveness of pre-screening before flaw detection is improved, and the condition of missing detection is effectively prevented.

The invention discloses an exoskeleton robot abnormal working condition automatic response and risk avoiding decision-making method, and relates to the technical field of exoskeleton, and the method comprises the steps: deploying a plurality of types of sensors to synchronously collect environment data, human physiological data and exoskeleton operation data, and carrying out the preprocessing to remove interference; through fusion of abnormal harmonic detection and a finite-state machine, dangerous working conditions such as wind speed abrupt change are identified, and grades are judged; monitoring eye movement and heart rate variability data in real time, quantifying cognitive load and grading; a decision rule base is constructed, and joint locking and other adaptive risk avoiding decisions are generated in combination with working condition types, grades and cognitive loads; the controller drives the execution mechanism to execute a decision, continuously monitors data at a frequency, dynamically adjusts the decision or recovers a normal mode. According to the method, dangerous working conditions are accurately identified through multi-source data fusion, adaptive risk avoiding measures are automatically triggered, a power assisting strategy is dynamically adjusted in combination with cognitive loads, and operation flexibility and physical output are balanced.

The invention provides an infrared physiotherapy cabin system and method based on animal and environment cooperative regulation and control, and relates to the field of animal physiotherapy and intelligent monitoring control. Comprising an animal state and cabin environment monitoring module, an infrared physiotherapy and cabin environment regulation and control module, an animal health assessment and curative effect assessment module, a personalized treatment and intelligent decision module and a man-machine interaction and safety supervision module. A credibility result is formed through multi-modal data synchronous alignment and data quality evaluation, health state characterization is constructed under credibility gating, a risk level is output, and infrared physiotherapy control parameters and cabin environment regulation and control parameters are generated through cooperative control under the constraint of a stage control boundary and a treatment target and are issued and executed. Continuously adjustable parameters are subjected to online fine adjustment in the physiotherapy process, and rollback and interlocking protection is executed when the risk is increased or the credibility is reduced; after the period is finished, the curative effect is evaluated, data recording and tracing archiving are completed, and collaborative closed-loop infrared physiotherapy regulation and control of the animal state and the in-cabin environment are achieved.

The invention belongs to the technical field of thermal power generation intelligent control, and discloses a boiler heating surface soot blowing opportunity judgment method and device, equipment and a medium, an input signal is constructed according to the load change rate, a compensation coefficient is output through a trained correction model, a boiler heating surface cleaning factor is compensated, and a soot blowing opportunity judgment result is obtained. Identifying that the boiler is in a statistical controlled stable operation interval, collecting historical measurement values of the compensation cleaning factor under various working conditions, and determining a safe operation lower limit; performing sliding window filtering processing on the current measurement value of the compensation cleaning factor to obtain a target cleaning factor; and if the target cleaning factor is continuously lower than the safe operation lower limit and exceeds the set duration, generating a soot blowing trigger instruction, thereby realizing dynamic compensation of the cleaning factor in a rapid variable load state and accurate screening of the stable operation state of the boiler, reducing misjudgment of the soot blowing opportunity, improving the accuracy of determination of the soot blowing opportunity of the heating surface of the boiler, and improving the reliability of the soot blowing opportunity. And intelligent control over soot blowing operation is improved.

The invention belongs to the technical field of document analysis, and particularly discloses a document analysis method and system, a terminal and a medium. Comprising the steps of obtaining content and page representation of a to-be-analyzed document; performing adaptive segmentation on the document content to obtain a segmentation sequence; performing joint coding on the text features, the visual features and the layout features of the segmented sequence to generate fusion feature representation used for representing cross-modal relevance; determining a hierarchical structure relationship of the document based on the fused feature representation; inputting a text corresponding to the hierarchical structure into a language model subjected to field training to generate a semantic analysis result of the document; and according to a semantic analysis result, generating knowledge representation of the documents and establishing associated information between the documents. According to the method, a consistent processing link can be formed among multi-modal feature fusion, document structure inference and semantic understanding, and high-precision structured analysis of complex document contents is realized.

The invention provides a data analysis system and method for evaluating a data center cold water link state, and belongs to the technical field of data center infrastructure intelligent operation and maintenance. According to the system, multi-source operation data of a cold machine, a water pump, a tail end air conditioner and a cabinet monitoring system in a cold water link are collected, and a dynamic threshold pool is constructed after working condition layering preprocessing so as to identify equipment abnormity; the method comprises the following steps: establishing a directed graph model based on a chilled water system topological structure, constructing a Bayesian network by fusing process logic, and reversely calculating a root cause posterior probability of each device when a cabinet hotspot event occurs; and generating an influence weight ranking list by integrating the abnormal strength of the equipment, the root cause probability, the topological key coefficient and the redundancy weakening coefficient, and further performing quantitative scoring and state grading on the overall availability of the cold water link. According to the method, working condition self-adaptive anomaly detection, robust root cause positioning and executable operation and maintenance priority ranking of the cold water link are realized, and the operation and maintenance efficiency and reliability of the refrigeration system of the data center are improved.

The application discloses a kind of hybrid power system health management method and system, belong to power system health management field.This application selects the characteristic signal of engine, battery and vehicle in hybrid power system to establish hybrid power system health degree;When hybrid power system fails, by obtaining characteristic signal and carrying out normalization processing, hybrid power system health degree based on grey correlation analysis method GRA is calculated in real time, and the misjudgment and the missed judgment of fault are reduced by the fault confirmation method based on counting;When hybrid power system fault is confirmed, through the power redistribution health management strategy under system fault based on LSTM-MPC, the change of hybrid power system speed, battery temperature and diesel engine exhaust temperature in limited time domain is predicted, and the predictive maintenance of hybrid power system health degree is carried out in combination with the change prediction result of part of the state of hybrid power system, so that hybrid power system can still operate with fault under the condition of meeting constraint.

The application discloses a colorectal cancer metastasis evaluation method based on NF-kappa B pathway activity, relates to the technical field of medical treatment, and comprises the following steps: obtaining a colorectal cancer tissue sample and a paired peripheral blood sample, separating exosomes, and dividing the tissue sample into a spatial resolution sub-sample and a function verification sub-sample; performing multiple immunofluorescence imaging on the spatial resolution sub-sample, identifying a phosphorylated p65 positive cancer cell group and a phosphorylated STAT3 positive immune cell group, and extracting cell phenotype data; calculating the spatial interaction weight value between high-activity cells based on the cell phenotype data to obtain a tumor microenvironment paracrine activation index; detecting the abundance of an upstream phosphorylated protein of the NF-kappa B pathway and the expression amount of a downstream metastasis effect gene in the function verification sub-sample, and obtaining a pathway function coupling score in combination with the matching degree of the two; and determining the content of a phosphorylated IKK protein and the content of a specific metastasis-promoting microRNA in the exosomes, and obtaining an exosome double-lock activity score by using the product relationship between the two.

The utility model provides a kind of piercing penetration induction device of electric spark puncher, including the adjusting support of installation on the machine table of electric spark puncher, ultraviolet light sensor device is fixed on the adjusting support, the ultraviolet light sensor device is located the side of the metal plate body to be perforated and points to its below, when perforating copper pipe passes through the metal plate body to be perforated, ultraviolet light sensor device senses the ultraviolet of the lower part perforation spark of the metal plate body to be perforated, and signal is transmitted to control system, the device is by ultraviolet sensor setting in the side of the metal plate body to be perforated, by the ultraviolet of the ultraviolet light sensor device after inductive electric spark passing through plate body to copper pipe return signal, the inductive wavelength of ultraviolet is between 185~260 nanometers, can induct the weak ultraviolet of electric spark generation, and setting in the side below of metal plate body, water and spark generated after copper pipe perforation cannot influence inductive device, delay service life and reduce misjudgment phenomenon.

The invention discloses a puncture tissue recognition and transmembrane event control method and system based on force sense and image fusion, and the method comprises the steps: collecting a force signal at the tail end of a puncture needle, a real-time ultrasonic image sequence and the pose data of a mechanical arm, and carrying out the timestamp alignment to construct a synchronous data set; on the basis, tissue stiffness, a force change rate, extreme value features, texture features, envelope boundary features and optical flow displacement features are extracted, the tissue level where the puncture needle is located is recognized through multi-modal fusion, and a membrane penetrating event is detected in combination with the force sense extreme value features and envelope boundary fracture; according to the tissue level and the transmembrane event, event driving control strategies such as constant-speed propulsion, deceleration early warning, rotation assistance, transmembrane braking and path compensation are automatically triggered, so that the puncture process is accurately regulated and controlled. The accuracy of puncture tissue recognition and the reliability of transmembrane event judgment can be remarkably improved, and the safety and stability of puncture operation are improved.

This application discloses a drone cargo delivery method, apparatus, device, and storage medium based on identity verification, applicable to drones equipped with sensors and cameras, and relating to the field of computer technology. The method includes: sending a random receiving code corresponding to the recipient's receiving location to receiving software; capturing a flash image of the recipient using the camera, detecting ambient brightness based on sensors, and adaptively processing the brightness of the flash image based on the ambient brightness; determining the flash brightness components based on the processed image using a target smoothing method, statistically analyzing the flash brightness components of each candidate region in the processed image, determining the target on / off state sequence corresponding to the flash based on the obtained flash energy sequence, and if the target on / off state sequence matches a preset on / off rule in the random receiving code, it indicates successful identity verification of the recipient, completing the corresponding cargo delivery operation. This improves the efficiency and accuracy of drone cargo delivery.

The application discloses a vehicle seat occupancy detection method and system and a storage medium, and relates to the technical field of vehicle intelligent detection. The method comprises the following steps: in response to an opening operation of a target vehicle door, camera sensing data in a preset time period is acquired, wherein the camera sensing data is used to represent vital sign object information sensed by a surround-view camera of a target vehicle within a preset range of the target vehicle door; and based on the camera sensing data and seat pressure sensor data, the occupancy of a vehicle seat corresponding to the target vehicle door is determined, wherein the seat pressure sensor data is used to represent force information of the vehicle seat sensed by a seat pressure sensor of the target vehicle. The application solves the technical problem that the accuracy of vehicle seat occupancy detection in the related art is low, and thus the vehicle safety reminder function cannot be effectively triggered to ensure passenger safety.

The invention provides a face recognition model training method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring a plurality of face image samples, and inputting each face image sample into a pre-trained teacher model and a to-be-trained student model to obtain teacher features, student features and prediction categories of each face image sample; determining feature distribution loss information based on the teacher features and the student features of each face image sample, and determining feature classification loss information based on the student features and the prediction categories of each face image sample; total loss information is determined based on the feature distribution loss information and the feature classification loss information; and based on the total loss information, training a to-be-trained student model until a preset training condition is met, and obtaining a final student model as a face recognition model. By adopting the method, the recognition accuracy of the face recognition model can be effectively improved on the premise of considering efficiency and computing resources.

The invention discloses a household appliance control method and a related device, and relates to the technical field of smart home / smart home, and the method comprises the steps: carrying out the fusion of data which is collected by a non-visual sensor and represents a first human body behavior, and data which represents a second human body behavior and represents an environment feature according to a data fusion strategy, and obtaining human body characteristic data representing a human body state. On the basis, an intention recognition model is called to conduct behavior intention recognition processing on the human body feature data, and recognition result data representing expected operation of the user on the associated equipment in the current environment is obtained. And finally, according to an equipment control instruction corresponding to the identification result data, controlling the household appliances associated with the identification result data. Through multi-sensor information fusion, the defects of a single sensor are effectively overcome, the probability of misjudgment and missed judgment is greatly reduced, dependence on a visual sensor is avoided, personal information safety of a user is protected, and the intention of the user can be more accurately understood.

The invention relates to the technical field of power supply ripple monitoring, in particular to a hydrogen production power supply ripple monitoring method for a water electrolysis hydrogen production system, and aims to solve the technical problem that the overall monitoring accuracy is insufficient due to the fact that a ripple coefficient calculated in the prior art cannot truly reflect the actual ripple condition. The method comprises the following steps: acquiring an output voltage signal and an input voltage signal of a hydrogen production power supply in a monitoring time period; extracting a direct current component and a plurality of alternating current mode components from the output voltage signal; identifying a ripple modal component from the plurality of AC modal components based on the periodic characteristics of each AC modal component and the time-frequency domain similarity between each AC modal component and the signal component of the corresponding frequency in the AC component of the input voltage signal; determining a ripple coefficient of the hydrogen production power supply according to the direct current component and the ripple modal component; and generating a ripple monitoring result of the hydrogen production power supply according to the ripple coefficient.

This application relates to the field of industrial data processing and intelligent control technology, and discloses a method for dynamic modeling and prediction of industrial big data combined with digital twins. The method includes: acquiring operating status data, process execution data, control command records and prediction results, and constructing a sequence of control action events; segmenting the operating status data according to the sequence of control action events to form natural operating data segments and intervention operating data segments, extracting distribution characteristics, fluctuation characteristics and response characteristics, and generating a segmented behavioral feature set; calculating the control reflexive influence quantity and generating a reflexive drift marker, dividing the natural evolution dataset and the control feedback dataset, and predicting them separately; generating control risk constraint quantity, screening, delaying or limiting the control commands to be executed, and iteratively correcting the dataset according to the execution results; this method can distinguish between control feedback changes and natural evolution changes, reduce drift misjudgments, and suppress multi-round oscillations and parameter fluctuations.

A kind of battery internal short-circuit nondestructive testing method based on magnetic field gradient distribution, it is related to battery technical field, steps include: in the state of battery constant current charging, constant current discharging or in the process of shelving, the external magnetic field distribution of measured battery is detected at any time interval using magnetic sensor;Obtain the relative change distribution of external magnetic field of battery in different time points under corresponding working condition state;The gradient distribution of magnetic field component in y and x direction of the relative change distribution of external magnetic field is calculated, and if the gradient distribution of magnetic field component is all in the same area Gradient sudden change appears, and all along the axis direction Gradient direction reversal appears, then it can be judged that internal short-circuit occurs in this area of the battery, the present application can accurately determine internal short-circuit fault, and can determine internal short-circuit site and internal short-circuit severity.

The utility model discloses an automatic detection stacking conveyor line, and relates to the technical field of stacking conveyor lines, the automatic detection stacking conveyor line comprises a feeding guide rail, one side of the feeding guide rail is provided with an opening, the opening is fixedly connected with a detection sorting reversing arc table, the input end of the detection sorting reversing arc table is in butt joint with the opening, and the output end of the detection sorting reversing arc table is in butt joint with the opening. The output end of the detecting, sorting and reversing arc table is fixedly connected with a material returning guide rail, the material returning guide rail and the feeding guide rail are arranged in parallel, one side of the detecting, sorting and reversing arc table is fixedly connected with a supporting plate, the supporting plate is located between the material returning guide rail and the feeding guide rail, and the supporting plate is rotationally connected with a rotating shaft; according to the utility model, the servo motor is used for accurately controlling the rotation of the material shifting hand, unqualified materials can be accurately shifted from the feeding guide rail to the material returning guide rail, the sorting accuracy and reliability are improved, the unqualified materials are transferred to a reinspection area through the material returning guide rail, the further confirmation of the product quality is facilitated, the misjudgment is reduced, and the production efficiency is improved. And the product quality control level is improved.

The invention relates to the technical field of automatic control, and discloses an operation data analysis and control method of intelligent cooking equipment, which comprises the following steps: continuously acquiring an operation state index of the cooking equipment, and acquiring a reference state parameter of a current cooking stage; constructing an operation state track of the cooking equipment based on the operation state indexes, and constructing a reference state track of the current cooking stage based on the reference state parameters; based on the running state track and the reference state track, whether a cooking mismatching state exists in the current cooking stage or not is recognized; if yes, feedback adjustment is carried out on the operation parameters of the cooking equipment until the cooking state regression condition is met; and otherwise, keeping the operation parameter as the preset value of the current cooking stage until the current cooking stage is finished. According to the method and the device, the running state of the cooking process is dynamically analyzed and fed back and adjusted, so that the stability of the cooking process and the consistency of cooking results are improved.

The invention provides a traffic signal lamp redundancy identification system and method based on red light narrowband filtering, and belongs to the technical field of intelligent driving environment perception. The system is characterized in that a main processing module processes a color image based on a YOLOv8n model and a lightweight convolutional neural network, and outputs a first recognition result and the confidence of the first recognition result; when the confidence coefficient is lower than a threshold value, a redundancy identification module is started, a narrow-band image obtained by an auxiliary imaging module is analyzed, whether a strong red light signal emitted by a traffic signal red light exists or not is judged through self-adaptive threshold value segmentation, morphological processing and feature analysis, and a second identification result is generated; and the decision fusion module fuses the two paths of results to generate a final control instruction. According to the invention, through physical spectrum filtering and double-path information fusion, interference of ambient light such as sunset and the like is effectively overcome, and the robustness and safety of an automatic driving system in recognizing traffic lights under complex illumination, especially red light, are remarkably improved.

The utility model discloses a high detection rate's stud mistake and miss detection equipment relates to stud detection technical field, including equipment main part, the upper side of fixed disc is connected with the ring sleeve that slides, and the deflection mechanism is connected between ring sleeve and fixed disc, and the outside of ring sleeve is provided with four groups of symmetrical arrangement's connecting groove, and the inboard fixed connection of ring sleeve has the baffle, and the upper side of equipment main part is connected with the feeding mechanism, and the upper side fixed connection of equipment main part has the vertical board, and the detection mechanism is installed outside vertical board, and the downside of fixed disc is provided with first blanking groove and second blanking groove, and the upside of fixed disc is provided with the round groove, and the inside of round groove is installed the plugging mechanism. This equipment adopts annular rotary detection station, and cooperates adjustable blanking channel, realizes workpiece dynamic multi -angle vision detection, and through real -time image intelligent determination stud quality, and utilizes rotary plugging mechanism switching blanking path, and automatically separates qualified and defective workpiece, and significantly improves detection accuracy and sorting efficiency.

This invention belongs to the field of power monitoring technology and discloses a brightness balancing and regional smoothing fusion algorithm for power monitoring images. The method includes: image denoising, using mean filtering and median filtering to smooth the image; image registration, performing optimal matching on two or more images obtained from different sensors, at different times, or from different angles; and image region fusion, smoothly fusing images with different angles and brightness contrasts. This method is applicable to panoramic video monitoring systems. Designed specifically for the complex lighting and high electromagnetic interference characteristics of power operation sites, this invention effectively eliminates on-site noise through adaptive preprocessing, and achieves a balance between global brightness uniformity and local detail preservation through layered brightness balancing, solving the problems of incomplete brightness balancing and detail loss in power scenarios encountered by general algorithms.

The application discloses a photovoltaic power station monitoring method and system based on three-dimensional image modeling and virtual simulation fusion, relates to the technical field of image modeling, and comprises the following steps: generating a reflection response graph according to an original inspection graph of a photovoltaic power station, sequentially performing adaptive threshold segmentation and morphological processing on the reflection response graph to obtain a candidate reflection graph of a screen; performing skeleton extraction on the candidate reflection graph of the screen to obtain a candidate highlight skeleton graph, screening out attached edge pixel points according to a skeleton pixel point set in the candidate highlight skeleton graph, connecting the attached edge pixel points to obtain an edge traction sliding graph; generating a basic outer adjacent band according to a three-dimensional model of a station of the photovoltaic power station, performing region merging on the basic outer adjacent band to obtain a grounding conductor layout band, and performing projection on the grounding conductor layout band to obtain a virtual screen corridor projection graph; and the application improves the reliability of a photovoltaic power station monitoring result.

The application discloses a kind of wild fire smoke cluster and its boundary vector automatic identification method, system, storage medium and electronic equipment, belong to satellite remote sensing and disaster monitoring technical field.Method includes: obtaining satellite active fire point, fire event and UVAI data;Fire source area is positioned based on fire event geographic location and initial plume pixel is marked;The spatially continuous smoke cluster connected domain is identified by iterative region growing algorithm;According to the specific distribution form of connected domain pixel in satellite pixel matrix, the accurate closed boundary vector is automatically generated using matching geometric algorithm;Finally, using multiple criteria, the spatial isolated independent smoke cluster is screened out.The application avoids the defects that traditional image method is easily disturbed by environment from physical principle, realizes fully automated processing without training data, and can output vector boundary that can be directly used for diffusion and emission evaluation, significantly improves the accuracy, automation degree and application value of wild fire smoke cluster monitoring.

This invention provides a method and system for remote meter reading and water usage information management for smart water meters. The method mainly includes: receiving raw data frames containing information such as cumulative flow values ​​and collection timestamps, and performing multi-dimensional quality checks to generate collection status identifiers; dividing the water usage interval into valid collection intervals and pending confirmation intervals based on the identifiers; calculating the measured water usage in the valid intervals and extracting abnormal features to generate initial abnormal events; generating a completion sequence for the pending confirmation intervals based on boundary flow values, historical water usage characteristics, and regional balance constraints; determining a supplementary reading control strategy and issuing supplementary reading instructions based on the completion sequence and other information; performing secondary verification of the initial abnormal events based on the supplementary data, and outputting the final abnormal result and meter reading data. This method improves the accuracy of water usage anomaly detection and the reliability of meter reading data through data quality grading, interval hierarchical management, and closed-loop verification of abnormal events.

The invention relates to the technical field of medical literature data processing and artificial intelligence, and discloses a random contrast test (RCT) bias risk automatic evaluation method and system, and the method comprises the steps: obtaining RCT literature data containing a text, a table and a flow chart, and a RoB2 evaluation problem set; a problem reconstruction module injected through medical knowledge; performing position decision and multi-modal coordination analysis on the RCT literature through a multi-modal document analysis module; the reconstructed evaluation problem and the analyzed context information are input into a multi-expert cooperation module, cooperative reasoning is performed through a multi-reasoning unit, results are fused through a decision unit, and item-level, field-level and overall bias risk evaluation results are generated. The invention also constructs bias risk entry level and domain level data sets for supporting the training and verification of the method. The system comprises a data acquisition module, a problem reconstruction module, a document analysis module and a collaborative evaluation module, all the modules work cooperatively, automatic evaluation of RoB2 whole-process bias risks is achieved, and the system can be applied to scenes such as evidence-based medical research, system evaluation and clinical research quality control.

The invention discloses a hematuria analysis method, device and equipment based on artificial intelligence and a storage medium. The method comprises the following steps: acquiring a urine visible component graph of urine to be analyzed and an interference state of the urine to be analyzed; selecting a hematuria analysis network model corresponding to the interference state from a preset corresponding relation set of the interference state and the hematuria analysis network model; the urine visible component graph is input into the hematuria analysis network model, the number of red blood cells corresponding to the to-be-analyzed urine is determined through the urine visible component graph, the hematuria category corresponding to the to-be-analyzed urine is determined based on the number of red blood cells, and the hematuria category comprises pseudo hematuria or true hematuria. According to the interference state of the to-be-analyzed urine, by constructing a plurality of corresponding hematuria analysis network models, misjudgment of the red blood cells in the to-be-analyzed urine can be reduced, accurate counting of the red blood cells is achieved, and therefore the accuracy of true hematuria detection is improved.

This application discloses an anomaly identification method, apparatus, and storage medium based on time series features, used to reduce misjudgments of abnormal moments in mine roadway space. The method involves: acquiring rock mass structure data and rock mass environment data of the mine roadway space; generating interlayer stability parameters based on the rock mass structure data and rock mass environment data; generating a window data matrix using the collected data set of mine characteristic indicators; generating an interlayer crack disturbance index set based on the collected data corresponding to the interlayer stability parameters and vibration-type features, and replacing the interlayer crack disturbance index set in the window data matrix; rearranging the elements of structural stability features and mine environment features in the window data matrix according to the interlayer crack disturbance index set; generating anomaly weights for each type of mine characteristic indicator in the window data matrix; generating an anomaly weighting matrix based on the anomaly weights and the window data matrix; and determining the anomaly occurrence time in the mine roadway space based on the anomaly weighting matrix.

The application relates to the technical field of agricultural and forestry product quality grading, and discloses a multi-modal agricultural and forestry product quality grading method based on cross-attention frequency perception Transformer (CAFFT), which comprises the following steps: obtaining an image of an agricultural and forestry product to be graded and performing pretreatment; generating a text description according to the image by using a multi-modal large language model; inputting the pretreated image into a visual Transformer encoder to extract visual features, and extracting intermediate visual features from an intermediate layer; respectively enhancing high-frequency detailed information and low-frequency global information through a frequency perception feature fusion module, and obtaining enhanced visual features through adaptive fusion; inputting the text description into a pre-trained language model to obtain text semantic features; performing bidirectional interaction and fusion on the enhanced visual features and the text semantic features through a multi-head complementary attention module to obtain multi-modal fusion features; and outputting the maturity degree, defect type or quality grade of the agricultural and forestry product by a classifier. The application can improve the accuracy and robustness of agricultural and forestry product quality grading.

The invention provides a radar point cloud-based foreign object intrusion detection method and system in a working area, and the method comprises the steps: setting a detection area range according to the working space of a mechanical connecting rod, and dynamically generating a direction bounding box (OBB) of the connecting rod through obtaining the posture state information and size data of the working connecting rod in real time; spatial screening is carried out on original radar point cloud data, only point cloud data which are located in a detection area and are not located in an OBB are reserved, and background point clouds outside the detection area and point clouds generated by a connecting rod are filtered out; and performing multi-dimensional anomaly detection on the filtered point cloud data, and comprehensively judging whether a continuously existing foreign object intrusion event exists or not. Through a linkage mechanism of point cloud filtering, background modeling and clustering verification, high-precision, low-false-alarm and real-time industrial intrusion detection is realized, and the problems of poor adaptability, high false alarm rate and the like of a traditional method in a complex environment are solved.