1950 results about "Heat map" patented technology

The invention discloses a power transmission line fault diagnosis and operation and maintenance scheduling method and system based on networking learning, and relates to the technical field of power transmission line fault diagnosis operation and maintenance scheduling. Related data is extracted to construct a high-risk equipment area and a visual high-risk area thermodynamic diagram, a visual risk grading diagram is constructed in combination with electrical quantity data, and meanwhile, an intelligent recognition storage network and a fault type classification recognition model are constructed in combination with a convolutional neural network-long and short-term memory network hybrid model; the model is optimized through networking learning and an attention mechanism, maintenance teams and resources are autonomously allocated in combination with an operation and maintenance management system, then autonomous optimization and closed-loop operation are achieved, full-process coverage of fault sensing, intelligent decision making and efficient response is achieved, the response time after a line fault occurs is remarkably shortened, and the maintenance efficiency is improved. And the fault handling and operation maintenance capabilities of the power grid system are comprehensively enhanced.

The invention discloses a forest steppe fire risk assessment method and system, and belongs to the technical field of forest steppe fire prevention. The problem of high false alarm and missing report rate caused by lagging fire risk identification, weak multi-source data fusion and insufficient dynamic response in the prior art is solved. According to the technical scheme, the method comprises the following steps: deploying a ground sensing node array to obtain real-time environment data of a grassland region, fusing high-resolution satellite remote sensing and regional weather forecast data, and constructing a space-time aligned risk assessment data cube; establishing a grassland fire risk factor dynamic coupling model, and dynamically allocating factor weights in combination with the adaptive weight decision tree; generating a comprehensive risk index, carrying out nonlinear mapping to five risk levels, and linking a visual engine to generate a dynamic thermodynamic diagram and push the dynamic thermodynamic diagram to a command terminal; the system supports online updating of the model, and weight parameters are automatically optimized based on a new fire event. According to the invention, high-precision, real-time and spatialized evaluation is realized, the early warning capability and prevention and control decision efficiency are improved, and ecological and economic losses are reduced.

The invention discloses a virtual power plant intelligent regulation and control method and system based on artificial intelligence, and belongs to the technical field of electric power system intelligent regulation and control, and the virtual power plant intelligent regulation and control method based on artificial intelligence comprises the following steps: S1, aggregating equipment side data, desensitizing to generate topological codes, and constructing time scale matrix synchronization; s2, constructing a dynamic model by equipment parameters, and mapping real-time data to output a difference map; s3, adding equipment constraints, building a multi-objective function, optimizing a strategy and performing correlation analysis; s4, a wind and light fluctuation overrun trigger RL strategy and an abnormal switching base line generate a mixed instruction; s5, locally verifying the instruction, and correcting and feeding back parameters if the prediction is out of limit; s6, generating a three-dimensional thermodynamic diagram, and displaying an association report and a historical record by AR; s7, aggregating the data to reconstruct the training set, locally fine-tuning the strategy network and performing incremental updating; the method has the beneficial effects that the regulation and control pain point of the virtual power plant is systematically solved, the operation and maintenance cost is reduced, the new energy consumption capability is improved, and the equipment out-of-limit risk is reduced.

The invention relates to an aeration fan predictive maintenance method, system and equipment based on multi-modal perception and adaptive learning and a medium. The method comprises the following steps: generating a time sequence data set through synchronous acquisition and combined noise reduction processing of a sensor group; generating a multi-dimensional feature vector through time-frequency feature spectrum characterization and interpretability contribution analysis in combination with dynamic weight distribution coupled by environmental factors; on the basis of the multi-dimensional feature vectors, real-time anomaly detection is carried out at the edge end through a lightweight model, and abnormal data fragments are uploaded to the cloud end; and performing cross-sensor bidirectional reasoning on abnormal data fragments through a reasoning model deployed at the cloud, reconstructing a sensor topological graph, intelligently triggering elastic incremental learning, cooperatively processing equipment degradation trend analysis, multi-source evidence fusion and space calibration, and outputting a life prediction result and a fault thermodynamic diagram. According to the method, the core pain points of high early fault omission ratio, insufficient model robustness and the like are solved, and cost reduction, efficiency improvement and equipment life prolonging are realized while the diagnosis precision is maintained.

The invention discloses a crop disease diffusion prediction method and system based on multi-modal fusion, and the method comprises the following steps: S1, collecting and preprocessing an RGB image sequence and a sensor data sequence of a crop growth environment, and generating an RGB image time sequence difference result and a sensor difference result through time difference processing; s2, mapping the RGB image time sequence difference result and the sensor difference result to a shared time sequence space through a time alignment algorithm, and generating a sensor alignment result and an RGB alignment result; s3, an FD-ViT prediction model is constructed; inputting the sensor alignment result and the RGB alignment result into an FD-ViT prediction model for prediction, and generating a prediction result; and S4, generating a disease diffusion thermodynamic diagram and early warning information according to a prediction result. According to the method, RGB image data and sensor network data are fused, a Transform-based time sequence prediction model is constructed, and early recognition and diffusion trend prediction of crop diseases are realized.

The invention discloses an auxiliary film reading method and system based on artificial intelligence, and the method comprises the steps: 1, collecting a pathological WSI, an electronic medical record, detection data and equipment parameters, correcting the equipment difference through adaptive dyeing normalization, and constructing a structured data package associated with an ID-timestamp of a patient; 2, developing a dynamic branch CNN, migrating teacher model knowledge through knowledge distillation, and introducing federated learning; step 3, the edge generates a thermodynamic diagram to mark a suspicious area, and the cloud outputs a structured report; step 4, constructing a normal tissue feature space by the variational auto-encoder, detecting abnormal slices and triggering expert re-checking; a reverse automatic encoder generates a pseudo-health image to compare and position a pathological area, and dynamic weight adjustment balances the federal learning convergence speed; 5, integrating the thermodynamic diagram, the gene data and the clinical indexes by a three-dimensional platform, and supporting multi-dimensional superposition display; webGL realizes browser end rendering, and NLP automatically generates a report abstract marked with a key evidence chain and is in butt joint with an international diagnosis and treatment guide.

The invention belongs to the technical field of spacecraft pose estimation, and particularly relates to a cross-domain spacecraft pose estimation method based on mask self-distillation domain adaptation, and the method comprises the steps: 1, inputting a complete image, and employing a Faster R-CNN algorithm to position a spacecraft bounding box; the robustness of the model is improved by applying a track environment data enhancement strategy; and extracting a target ROI region as key point regression network input based on the detection frame. 2, dividing ROI (Region of Interest) data of a source domain and a target domain; and optimizing heat map supervision loss learning key point positioning knowledge. 3, inputting random mask enhanced target domain data into the student model; inputting original target domain data into the teacher model; a learnable shared prototype space is constructed, and self-distillation is guided through heat map consistency loss and semantic consistency loss. And 4, jointly optimizing the loss of the key point regression network 3, and realizing progressive migration of source domain annotation knowledge to a target domain. And 5, solving the 6D pose of the spacecraft relative to the camera through the EPnP. According to the invention, robust six-degree-of-freedom pose estimation of the target spacecraft is realized.

The invention discloses a zero sample anomaly detection method based on causal learning, and the method comprises the steps: collecting an input image sample, inputting the input image sample into a vision-language model, and carrying out the extraction and fusion of multi-scale features; a learnable text prompt vector is generated through a text encoder, the extracted multi-scale fusion features and the text prompt vector are input into a comparison learning core module, similarity is calculated through learnable prompts, and abnormal probability distribution and a thermodynamic diagram are generated; constructing a causal learning module by using the extracted multi-scale fusion features and a comparative learning result output by the comparative learning core module, correcting a similarity result through multi-scale semantic constraints, and calculating a causal gap between a normal region and an abnormal region; designing a multi-objective joint loss function, and performing training optimization on the vision-language model; and the vision-language model after training optimization is used for anomaly detection, and a detection result is output. The problem of insufficient utilization of vision-language model features is solved.

The invention relates to a power distribution network abnormal state intelligent identification method based on unmanned aerial vehicle routing inspection, and the method comprises the following steps: S1, obtaining power distribution network routing inspection related data, including asset main data, historical operation and maintenance data and environment constraints; s2, constructing the digital twinning of the power distribution network based on the power distribution network inspection related data, and obtaining the mapping from the equipment ID to the spatial pose and parameters; s3, generating a risk heat map based on historical defects and environmental risk scoring; s4, constructing an equipment-geography-working condition knowledge graph, and generating inspection targets and priorities according to line sections in combination with mapping from equipment IDs to spatial poses and parameters and a risk heat map to obtain a task package; and S5, performing multi-target route inspection according to the task package, the risk heat map and the mapping from the equipment ID to the spatial pose and the parameter, and intelligently identifying the abnormal state of the power distribution network. The operation reliability of the power distribution network is effectively improved, and the power supply quality is improved.

The invention belongs to the technical field of track detection, and more specifically relates to a track scene thunder-vision fusion anti-intrusion monitoring method and system based on sparse feature fusion. The method comprises the steps that a laser radar obtains three-dimensional point cloud data, a high-definition camera obtains visible light image data, and data preprocessing is carried out; performing external parameter automatic calibration optimization and time alignment compensation on the laser radar and the high-definition camera to obtain time-space aligned multi-modal data; fusing the camera semantic features and the laser radar geometric features based on multi-modal data of space-time alignment to obtain fused features; and after multi-modal feature fusion is completed, a target heat map is generated through a sparse detection head, and graded early warning is realized in combination with a dynamic safety distance model. The method solves the problems that an existing recognition algorithm is high in false alarm rate, and the false alarm rate is higher under low-light conditions such as rainy days or dusk; the data dependence is strong, and the ability to detect unlabeled novel foreign matters is almost zero.

The invention relates to the technical field of industrial visual inspection, and discloses a leather surface defect identification method and system based on image processing. The method comprises the steps of performing preprocessing and multi-scale feature extraction on a leather image, generating a defect thermodynamic diagram through an attention mechanism, and dividing candidate regions; innovatively constructing a regional association graph, updating node embedding by using a graph attention network, and fusing global context information; and generating instance segmentation results through adaptive clustering and classifying the instance segmentation results. According to the method, region relevance is modeled through a graph structure, so that recognition of each region is benefited from global context information, and the problem that segmentation of adhesion and irregular defects is inaccurate in a traditional method is effectively solved; a clustering type instance generation mechanism can automatically aggregate discrete regions according to deep feature similarity, so that the defect contour segmentation precision and instance distinguishing capability under a complex texture background are improved, and accurate conversion from pixel-level prediction to instance-level segmentation is realized.

The invention relates to the technical field of defect detection systems, and discloses a titanium alloy surface crack defect detection system based on deep learning. According to the system, a crack feature extraction module is used for collecting a titanium alloy surface image and extracting multi-scale crack features including crack trend distribution features and micro-crack density features; the multi-modal data fusion module is used for receiving the multi-scale crack features and performing space-time alignment on the multi-scale crack features and ultrasonic reflection wave features collected in real time to generate a fusion defect feature matrix; the dynamic learning engine module is used for constructing a crack propagation prediction model according to the historical change trend of the fusion defect feature matrix and outputting a dynamic defect response vector; the defect positioning module is used for mapping the dynamic defect response vector to a titanium alloy surface three-dimensional coordinate space to generate a defect position thermodynamic diagram; and the self-adaptive scanning control module is used for analyzing the defect confidence of each area in the defect position thermodynamic diagram and dynamically adjusting the scanning path and the focal length parameter of the industrial camera.

The invention discloses a lightweight display interface rendering optimization system, which relates to the technical field of image processing, and comprises an acquisition module, an interface analysis module, a behavior analysis module and an attention analysis and rendering module, dynamically calculating an effective view radius by utilizing a visual tunneling effect; the interface analysis module performs character string fuzzy matching in combination with the context input by the user, identifies the search intention of the user and improves the weight of a corresponding component; the attention analysis module performs multi-modal weighted fusion on the physiological fixation data and the sparse content saliency thermodynamic diagram; according to the method, by recognizing the semantic intention and the physiological fixation point of the user, on the premise that core visual experience continuity is guaranteed, GPU load and video memory bandwidth occupation are remarkably reduced, and balance of high-performance display and low-power-consumption operation is achieved.

The invention discloses a full waveform decomposition method based on a multi-branch convolutional neural network, which belongs to the technical field of airborne depth sounding laser radars, is used for airborne depth sounding laser radar echo signal decomposition, and comprises the following steps: obtaining and preprocessing an original echo sequence, constructing a multi-branch one-dimensional convolutional neural network model, and performing neural network training; and based on the trained multi-branch one-dimensional convolutional neural network, outputting a three-channel probability heat map, performing multi-peak sub-pixel decoding on the three-channel probability heat map, and outputting a peak accurate position, a peak accurate position normalized value and a peak confidence after de-weighting. According to the method, the multi-branch one-dimensional convolutional neural network model is constructed and multi-peak sub-pixel decoding is carried out, so that stable decomposition of full-waveform echoes and high-precision positioning of multi-echo peak values are realized under the conditions of multi-peak superposition, high noise and weak signals, and false detection and false peaks caused by noise are remarkably reduced.

To provide a combined navigation positioning method for extreme weather rescue.SOLUTION: A method disclosed herein comprises: a step 1 of constructing an original training data set; a step 2 of adding incremental data set training samples on the basis of the original training data set, constructing an incremental grid real-time correction model, and obtaining a real-time dynamically updated multipath heat map; and a step 3 of assisting GNSS / IMU tight combination solution according to the real-time dynamically updated multipath heat map, and completing the combined navigation and positioning for extreme weather rescue.SELECTED DRAWING: Figure 1

The invention relates to the technical field of pedestrian environment perception, and discloses a pedestrian floor tile environment perception method based on multi-modal fusion. The method comprises the following steps: acquiring a multi-mode sensing data sequence of pedestrian floor tiles in a preset monitoring area, wherein the multi-mode sensing data sequence comprises pressure distribution, a surface texture image and an environment temperature and humidity data sequence; then, combining a preset environmental parameter dynamic threshold value and target sensing area spatial characteristics, and screening a dynamic sensing target area through positive equilibrium relation matching, the former being pre-configured based on historical sensing data statistical distribution and a preset environmental anomaly judgment rule; the latter is calculated based on floor tile laying density, a pedestrian flow density peak value and an abnormal region thermodynamic diagram distribution range; dividing a plurality of continuous sensing time windows according to the boundary position of the target area and the multi-modal sensing node distribution topology; and finally, according to a window sequence, window sensing operation is completed in a gradual multi-source data fusion mode, and comprehensive sensing of the environment state of the pedestrian floor tiles is realized.

The invention discloses an intelligent fire-fighting AI collaborative early warning system and method fusing GIS and multi-source data, and relates to the technical field of fire-fighting safety, and the system comprises a GIS space modeling module which is used for constructing a digital twinborn model through integrating a building BIM model, fire-fighting facility distribution, a population density thermodynamic diagram, sensing equipment layout and real-time environment data; the multi-source data fusion module is used for integrating sensing equipment data, real-time environment data, meteorological information, manually reported hidden dangers and a historical fire case library, and outputting the data after standardization processing; the AI collaborative early warning engine module is used for accessing standardized data, realizing building fire dynamic risk assessment and outputting graded warning information; the 3D simulation drilling platform is used for calling the digital twinborn model, receiving graded alarm information and supporting fire spreading simulation, emergency drilling and plan optimization; and the block chain evidence storage module is used for storing alarm information and drill records. According to the invention, risk assessment and accurate prediction of the fire hazard can be realized.

The invention provides an automatic driving test scene generation method based on real traffic data, and solves the problems of low accident data utilization rate, SIL / HIL test splitting and insufficient boundary coverage in the prior art. Comprising the following steps: acquiring multi-source heterogeneous traffic accident data; cleaning data by adopting a joint interpolation-anomaly detection mechanism; vehicle dynamic sudden change characteristics within 0.5 second before braking are extracted through LSTM and DTW algorithms; constructing a three-dimensional scene pipeline driven by a physical engine, and dynamically associating the pavement slippery coefficient with the rainfall intensity; analyzing the accident text into simulation parameters by using a semantic-physical parameter converter; performing SIL-HIL cooperative verification: performing extreme illumination perception test and narrow road planning verification in an SIL environment, and realizing 1ms step length fault injection test in an HIL environment; positioning failure parameters based on Bayesian optimization; a GAN is adopted to generate a long-tail scene, and a test boundary is expanded by coupling extreme conditions such as rainstorm / low visibility; and outputting a standard scene library containing the collision probability thermodynamic diagram. The safety verification efficiency under the extreme working condition is remarkably improved.

This disclosure describes systems, methods, and devices related to object detection in images. A device may input an image, representing an object, to a manual labeling learner system; identify, using the system, first coordinates of an upper left corner of a bounding box representing the object based on a heatmap indicative of a probability of the first coordinates representing the upper left corner; identify, using the system, second coordinates of a bottom right corner of the bounding box based on the first coordinates and a first distance regression map indicative of coordinate differences between the second coordinates and ground truth coordinates input to the machine learning model as training data; generate, using the system, adjustments to the first coordinates and the second coordinates based on a second regression map; and generate, using the system, the adjusted first and second coordinates, the bounding box.

The invention relates to the technical field of multi-unmanned-aerial-vehicle cooperative operation and three-dimensional modeling, in particular to a multi-unmanned-aerial-vehicle cooperative three-dimensional rapid modeling method for a highway accident scene, and the method comprises the steps: generating a three-dimensional grid map of an accident area through the scanning of a millimeter-wave radar by a main control unmanned aerial vehicle; subareas are divided according to a load balancing strategy and are distributed to slave unmanned aerial vehicles, the slave unmanned aerial vehicles traverse grids along a snake-shaped track, laser radar point clouds and five-view-angle images are synchronously collected, data are bound through double time stamps and space coordinates, the point clouds are preprocessed through edge computing nodes, and the point clouds are stored in a database; the master control unmanned aerial vehicle evaluates quality based on density standard deviation and overlapping matching degree and instructs to reacquire, performs high-precision Poisson reconstruction on an accident core area, performs voxelization processing on a peripheral area, maps image textures, optimizes vehicle deformation details, simplifies a model and retains key element precision, and finally performs data processing. License plate coordinates, a scattered object thermodynamic diagram and an emergency lane occupation state are automatically marked, a visual model is generated, and rapid and accurate restoration of an accident scene is realized.

The invention relates to the field of water environment monitoring and treatment, in particular to a lake and reservoir pollution flux dynamic evaluation and optimization system and method.According to the system, hydrological, water flow, meteorological and remote sensing data are collected in real time through a data processing module, and a pollution flux space-time matrix is generated; the atlas generation module constructs a mechanism model based on the matrix and outputs a pollution flux thermodynamic atlas; the scheme simulation module determines a pollution source emission reduction scheme according to the map and simulates the influence of the pollution source emission reduction scheme on lake and reservoir water quality and ecology; and the feedback optimization module performs treatment efficiency scoring on the simulation result, compares the evaluation result with a preset threshold value, and adjusts the emission reduction scheme and updates the thermodynamic map when the evaluation result does not reach the standard. According to the method, through data driving, map visualization, scene simulation and dynamic feedback closed loop, accurate evaluation and optimal regulation and control of the lake and reservoir pollution flux are realized, and a scientific basis is provided for pollution abatement decision.

The invention relates to the technical field of building intelligent control and energy saving, and discloses a building equipment intelligent control optimization management system based on a green low-carbon building. The system comprises five subsystems, wherein an environmental parameter sensing subsystem is used for acquiring temperature and humidity gradients, illumination intensity distribution and personnel activity thermodynamic diagram data inside and outside a building through a distributed sensing network to form a multi-dimensional environmental state vector; the equipment operation state acquisition subsystem acquires an operation power curve, an energy efficiency conversion rate curve and an equipment health degree index of heating ventilation air conditioning, illumination and renewable energy equipment in real time; the dynamic load prediction subsystem establishes an energy consumption load space-time distribution prediction model according to historical data time sequence relevance; the control strategy generation subsystem is used for generating an optimized instruction set containing an equipment start-stop time sequence, a power regulation gradient and an energy distribution weight in combination with the current environment vector and a predicted load; the execution terminal adapts the subsystem to convert the instruction into a device compatible signal and calibrate a response delay.

The invention relates to the field of biological medicine, and discloses a method and system for analyzing cough sound by using disease characteristics to diagnose respiratory diseases, and the method comprises the steps: deploying a six-microphone annular array to achieve the precise positioning and triggering of a sound source; self-adaptive spectral subtraction and Wiener filtering cascade are adopted to enhance the audio; segmenting a cough segment based on energy envelope; fusing the Mel-cepstrum, the linear prediction residual error, the harmonic energy ratio and the transient zero-crossing rate to construct a pathological feature matrix; extracting local, medium-range and global time sequence features through a three-branch parallel convolutional network; inputting a disease specific classifier to discriminate asthma, pneumonia and laryngitis respectively, and applying a feature decoupling regular term to improve interpretability. The system correspondingly realizes the modularized processing flow. According to the method, the cough sound collection quality and the disease subtype recognition accuracy in a complex environment are improved, meanwhile, the thermodynamic diagram is output to assist clinical decision making, and the diagnosis credibility and practicability are enhanced.

The invention provides a geological environment monitoring method and system, and the method comprises the steps: constructing an air-space-earth-depth four-dimensional cooperative monitoring network, collecting multi-modal geological environment data, carrying out the preprocessing of the data, and obtaining multi-modal feature data and abnormal data points after dynamic load balance distribution of calculation resources; fusing the data to generate a comprehensive geological environment feature map containing space-time correlation features, abnormal hot spot distribution and a geologic body three-dimensional reconstruction model; and based on the map, performing geological risk prediction by using a geological risk prediction model to obtain a prediction probability. A monitoring network is constructed to integrate multi-modal geological data, resource allocation is optimized by combining edge calculation dynamic load balancing, geological risk prediction is realized by using a prediction model, the problems of single data, poor dynamic adaptability and weak model generalization ability of a traditional method are solved, the prediction precision is improved, the response time is shortened, and the prediction efficiency is improved. And a high-risk area is visually displayed through a three-dimensional risk thermodynamic diagram.

The invention belongs to the technical field of analog simulation, and particularly discloses an intelligent agent-based dynamic confrontation simulation system, which comprises a data acquisition module, an intelligent analysis module, a decision generation module, an intelligent agent behavior self-adaption module, a training evaluation module and a multi-mode man-machine interaction module, physiological, action, voice and eye movement data of trainees are collected in real time through a multi-modal sensor, and a tactical intention is recognized and a dynamic three-dimensional battlefield situation thermodynamic diagram is generated in combination with virtual battlefield environment parameters; an agent coping strategy is generated based on reinforcement learning and a decision tree, and an agent is driven to carry out real-time confrontation; and performing multi-dimensional quantitative evaluation on the whole training process through a training evaluation module, and performing closed-loop optimization on an agent decision and strategy library based on an evaluation result. The method supports various natural interaction modes such as voice, gestures and eye movement, remarkably improves the fidelity, intelligence and training efficiency of simulation training, and is suitable for the field of military training and tactical drilling.

The invention discloses a CT image analysis method and system based on a neural network, and relates to the technical field of CT image analys.The method comprises the steps that an original CT image is obtained after user authorization, a Laplace operator is adopted to strengthen a focus boundary, and a circular region of interest is intercepted to remove edge sensitive information; extracting edge and texture information in the standardized image; focus area features are focused step by step; executing characteristic distillation balance based on category sample distribution, and outputting a focus characteristic graph with local perception enhancement and sample balance characteristics; segmenting the lesion feature map into serialized units, embedding position codes, inputting the serialized units into a plurality of layers of encoders, and fusing an image structure and text indication information through a dynamic adjustment mechanism; performing linear classification on the global semantic vector to output a diagnosis result, generating a focus thermodynamic diagram, and superposing the focus thermodynamic diagram to an original image for visualization; and performing dynamic optimization based on doctor feedback. The accuracy of feature analysis is improved; the overall operation efficiency of the system is improved.

Aspects of the disclosed technology include computer-implemented systems and methods for machine-learned multimodal models for feedback predictions for synthetic content. A machine-learned multimodal model is configured to generate a feature map based at least in part on fusion of image information and text information from a synthetic image and a text prompt. The model is configured to generate a set of text tokens based at least in part on fusion of the image information and the text information. The model is configured to generate at least one misalignment or implausibility heatmap based at least in part on the at least one feature map. The model is configured to generate at least one predicted misalignment sequence based at least in part on the set of text tokens.

Aspects of the disclosed technology include computer-implemented systems and methods for machine-learned multimodal models. A machine-learned multimodal model includes one or more embedding layers configured to generate one or more image tokens and one or more text tokens in response to the imagery and the text, a transformer encoder configured to receive the one or more image tokens and the one or more text tokens and generate one or more fused image tokens and one or more fused text tokens, a heatmap predictor configured to obtain the one or more fused image tokens and generate at least one image heatmap, and a sequence predictor configured to obtain the one or more fused image tokens and the one or more fused text tokens and generate a predicted sequence associated with the image.

The invention discloses a data deep learning and intelligent analysis method based on an AI artificial intelligence technology, and relates to the technical field of basic AI models, and the method comprises the steps: employing a multi-modal data preprocessing module to carry out the expansion of small sample data through a generative model, and combining with meta-learning to extract prototype features, meanwhile, an epsilon-differential privacy budget is dynamically allocated based on the data sensitivity level so as to inject dynamic noise; establishing a layered federated learning architecture, training a model by local training nodes through a loss function containing a self-adaptive regularization item, and performing sparse processing and gradient disturbance before uploading parameters; the global aggregation node adopts a weighted federated average algorithm to aggregate parameters, and dynamically adjusts the communication frequency according to the loss convergence speed; and a target model is obtained through iterative training, and a decision interpretation report containing the attention thermodynamic diagram and the desensitization identifier is generated when a result is output. According to the method, the problems of small sample overfitting, data islands and privacy disclosure are effectively solved, and the accuracy and practicability of the model are improved.

The invention relates to the technical field of geological early warning, in particular to a dangerous rock mass instability analysis method, system and equipment based on a space-time diagram neural network, by fusing unmanned aerial vehicle LiDAR, multispectral data, meteorological radar data and the space-time diagram neural network (ST-GNN), the system realizes sub-meter spatial resolution and minute-level time response, and the stability of dangerous rock mass instability analysis is improved. The four-dimensional (time and space) analysis result of the instability probability of the dangerous rock mass is obtained through high-precision space-time modeling, the problems that a traditional geological disaster early warning system is low in resolution ratio, slow in response and high in misinformation are solved, the comprehensiveness, accuracy and reliability of instability prediction of the dangerous rock mass are improved, and the early warning effect is good. And full-chain intelligent closed-loop management of real-time data acquisition-dynamic prediction-early warning push-feedback optimization is supported, the emergency decision time is shortened by real-time rainfall superposition risk thermodynamic diagrams, and the attenuation rate of long-term prediction precision is reduced by dynamically fusing newly added geological data and instability events through incremental learning.