1696results about How to "Reduce computing cost" patented technology

The invention belongs to the technical field of image classification neural networks, and discloses a lightweight image classification neural network architecture system based on knowledge distillation. The invention provides a lightweight image classification neural network architecture system based on knowledge distillation. The lightweight image classification neural network architecture system comprises a data set preprocessing module, a teacher model preprocessing module, a subnet model construction module, a personalized training module, a real loss calculation module, a distillation loss calculation module, a total loss calculation module and a student model construction module. The invention provides a lightweight neural network architecture system based on knowledge distillation, and designs a brand new training mode by using the superiority of knowledge distillation for teacher model weight extraction: the knowledge distillation of teacher and student models is combined with a deep compression method; the system for constructing the lightweight network can be used for training various complex image classification neural networks so as to improve the network performance.

The invention relates to a PCB defect detection method and system based on a YOLOv8 network, and belongs to the technical field of target detection. Collecting and preprocessing a PCB sample image data set for training a PCB defect detection model; inputting a to-be-detected image into the trained defect detection model to obtain a detection result; the to-be-detected image is generated by collecting and processing image data of the to-be-detected PCB; wherein the defect detection model comprises a trunk network, a neck network and a head network, and the trunk network extracts features through a mobile reverse bottleneck convolution module and a C2f module to generate an intermediate feature map; and the neck network performs splicing fusion on the middle feature map to generate a spliced feature map, the head network performs re-weighting on the spliced feature map through a convolution kernel attention fusion module to generate a third feature map, and then the detection module outputs a detection result. The method not only improves the precision of the defect detection model in detecting the features of the tiny target, but also improves the detection speed and generalization ability, and reduces the calculation cost and complexity at the same time.

The invention discloses a mobile robot autonomous exploration method based on deep reinforcement learning, and the method comprises the steps: starting laser radar equipment, a calculation unit and a mobile robot body power supply in an unknown environment, and collecting the point cloud data of the current environment through a laser radar; storing the point cloud data based on the complementary hole structure, constructing a three-dimensional point cloud map, and constructing a two-dimensional grid occupation map after projection; constructing an information graph based on the two-dimensional grid-occupying map, and sparsifying the information graph by using a bidirectional A * algorithm to form a sparse information graph; a reinforcement learning strategy network is constructed, a target function of the reinforcement learning strategy network is designed, the input of the reinforcement learning strategy network is a sparse information graph, and the output of the reinforcement learning strategy network is a target path point at the next moment; a path planner generates a collision-free path according to the target path point, and a local planner generates a linear velocity value and an angular velocity value of the current robot according to the collision-free path; a motion controller of the robot generates a motion command according to the speed value, the mobile robot is controlled to reach the target path point, and the steps are repeated until the robot autonomous exploration task based on deep reinforcement learning is completed; the method has the advantages that a model-free deep reinforcement learning technology is adopted to learn an exploration strategy in a simulation system, the method can be seamlessly transplanted to a real physical environment, higher-efficiency autonomous exploration can be realized, the adaptability to the environment is high, the application scene is wider, and a foundation is laid for completely autonomous operation of the mobile robot.

The invention relates to the technical field of data processing, and discloses an intelligent question answering method and device fusing a knowledge graph and a large model, and a storage medium, and the method comprises the steps: obtaining a to-be-answered original question; performing entity extraction on the original problem to generate an entity information set, and performing problem rewriting to generate similar problems of the original problem; traversing each piece of entity information in the entity information set, and searching and screening in a preset knowledge graph based on a search strategy fusing similar problems to obtain a knowledge path; obtaining knowledge according to the knowledge path, and according to the task instruction and the original question, performing logical reasoning by combining an answer generation large model with the obtained knowledge to generate a prediction answer of the original question; and carrying out authenticity detection on the predicted answers, filtering the predicted answers, and generating target answers. According to the method, the possibility that a large model generates unreal or irrelevant information is reduced, the calculation cost is reduced, the illusion problem of generating answers by the large model is relieved, and the credibility and the accuracy of the answers are improved.

The invention provides a visible light and thermal infrared image target fusion detection method for the visual angle of an unmanned aerial vehicle, and belongs to the technical field of image processing, and the method comprises the steps: carrying out the preprocessing and data enhancement of a data set of visible light and thermal infrared images of a detection region obtained by the unmanned aerial vehicle, and obtaining a plurality of initial images; initial images under different scales are analyzed based on a multi-scale dynamic feature pyramid network, fusion processing is carried out on the initial image under a visible light mode and the initial image under a thermal infrared mode based on an attention fusion mechanism, a candidate area is locked, and then highlighting processing is carried out on a tiny target, a long-distance target and the candidate area; and splitting all the target samples to obtain a training set and a verification set, training a preset network model based on the training set to obtain an initial detection model, and performing model performance evaluation and model optimization on the initial detection model based on the verification set to obtain a target detection model. The problem of target detection challenge of the unmanned aerial vehicle in a complex environment is solved.

The invention belongs to the technical field of computer vision, and particularly provides an improved YOLOV8-based offshore wind turbine blade surface defect detection method, which comprises the following steps: constructing an offshore wind turbine blade defect data set, marking defects on an image, and constructing a training set, a verification set and a test set based on the data set; a YOLOv8 baseline network model is improved, a C2f-RepGhost lightweight module is introduced to replace a C2f module in a trunk feature extraction network, a BiFPN structure is used to improve the C2f module to achieve the purpose of feature fusion, an attention module CBAM is introduced, and a final defect detection model is constructed; training is carried out to optimize a network model of the YOLOv8; and inputting a program based on the trained weight file, and testing the detection speed and precision of the YOLOv8. The improved YOLOv8 algorithm can effectively improve the precision and speed of identifying the surface defects of the offshore wind turbine blade, reduces the parameter quantity of the model, reduces the calculation cost, has better performance compared with other mainstream algorithms, and can be applied to many places.

The invention discloses a lifelong target re-identification method based on descriptive text prompt prototype compensation, relates to the field of computer vision, and aims to display and solve the problem of feature drift caused by model increment update by introducing a prompt prototype compensation module. The module generates a multi-modal vision and text prototype for each target category through a vision language pre-training model, and maps old task features to a new task feature space through a cross attention mechanism so as to realize fine-grained semantic drift compensation. In addition, a boundary perception prototype correction module is designed to solve the problem of boundary confusion caused by updating of old prototypes. The module dynamically samples false features in compensation distribution to correct old prototypes with fuzzy boundaries, focuses on categories easy to confuse to maintain clear decision boundaries, optimizes cross-task and intra-task correction loss to reduce category confusion, effectively alleviates the problem of disastrous forgetting, and improves the adaptability and generalization ability of the model.

The invention provides a potential diffusion model-based building planning image generation method and system. The method comprises the steps of data preprocessing, potential diffusion model structure improvement, input condition optimization, model training, image generation and the like. The relevance between text description and architectural styles is enhanced by preprocessing data sets of architectural design images and urban planning images and introducing channel crossing and space crossing attention mechanisms. Meanwhile, an architectural style embedding vector, an architectural description text embedding vector and a geometric information embedding vector are combined as additional input conditions to ensure that the generated image contains complete information. In a model training process, a multi-scale strategy and depth separable convolution are adopted, so that the recognition capability and training efficiency of the model to a large scene are improved. The system comprises a data preprocessing module, a model optimization module, a model training module, an image generation module and the like, and can generate a high-resolution building image conforming to design specifications. The method and system can be widely applied to the fields of building design and urban planning, and design efficiency and accuracy are improved.

The invention discloses a multi-task blood vessel segmentation model construction method based on a hybrid encoder. The encoder is a Swin Transform path multi-layer encoder based on a Mamform module, and the Mamform multi-layer encoder carries out data processing in combination with a W-MSA and an improved Mamba module; the decoder is a multi-task decoder and consists of three independent decoding paths which are respectively used for blood vessel segmentation, contour edge segmentation and blood vessel connectivity segmentation, and each decoding path adopts Swin Transformblock as a construction block and is used for generating feature mapping with the same size as an input image; an attention mechanism is shared, learned features are shared among decoders, the number of model parameters can be reduced, and overfitting is prevented; in the training process, a total loss function is formed through a loss function of a vascular connectivity decoding path, a self-distillation loss function and the like, the total loss function is calculated, model parameters are updated by using a back propagation algorithm, the segmentation performance of the model is gradually optimized, and after training is completed, the model can carry out precision on a vascular image.

The invention discloses a weighing sensor test compensation method based on fuzzy recognition, and relates to the technical field of sensor test compensation, and the method comprises the following compensation steps: inputting an original reading of a weighing sensor into a calculation module, and outputting a final compensated reading by the calculation module. According to the invention, the influence of various environmental factors such as ground flatness, power supply stability, temperature, humidity, air pressure, altitude and the like on reading is comprehensively considered, preliminary compensation and comprehensive compensation based on position and time are realized, and the adaptability and measurement precision of the weighing sensor under different environmental conditions are remarkably improved; a vibration correction coefficient is introduced to effectively compensate the influence of vibration on reading, the direct influence of vibration strength is considered, the compensation strength is dynamically adjusted through a modulation factor, the compensation effect is more accurate and effective, automatic adjustment and optimization of compensation parameters can be achieved, and the method is suitable for large-scale popularization and application. Therefore, the weighing sensor can automatically adapt to a complex and changeable measurement environment.

The invention discloses a feature classifier-based thinking chain reasoning mathematical problem solving method and system. The method comprises the following steps of: (1) acquiring a mathematical problem set, and selecting forward and reverse thinking chain examples to splice to form a new set; (2) respectively inputting the new set elements into a model for reasoning to construct a word-level reasoning path spanning tree, and selecting an attention weight matrix after pooling difference processing as node features for storage; (3) traversing all the generated reasoning path spanning trees, and screening nodes meeting requirements to construct a feature classifier training set; (4) training a feature classifier by using a support vector machine algorithm; and (5) through the trained feature classifier, participating in path selection in a pre-training language model reasoning process, and obtaining a relatively accurate reasoning process and answers. By means of the method, adjustment and control of finer granularity of the reasoning path of the pre-training language model can be achieved, accurate reasoning solving of the pre-training language model on mathematical problems is facilitated, and the generalization level of the pre-training language model is improved.

The invention relates to the technical field of multi-round dialogue processing, in particular to a multi-round dialogue processing method based on RAG technology dynamic cue word retrieval. Comprising the following steps of question input by a user, initial retrieval and cue word generation, question splicing and answer generation, historical dialogue management and context coding, multi-round question analysis and retrieval enhancement, complex question decomposition and parallel retrieval, multi-source information fusion and generation model optimization, iterative feedback and model tuning, and exception handling and fault-tolerant mechanism. According to the method, related information is retrieved from an external knowledge base, the model can utilize wider and more specific knowledge to generate answers and better track dialogue history and context, so that answers consistent and coherent with the dialogue history are generated, the model can generate answers on a short input sequence by retrieving and splicing related cue words, and meanwhile, the model can be used for analyzing the dialogue history. Cue words are dynamically generated according to user input and retrieval results, so that the model can better adapt to different dialogue scenes and user requirements.

The invention discloses a train key component online health monitoring method based on multi-sensor signal cooperation, and the method comprises the steps: selecting an indirect signal as an edge end train online abnormal state early warning index based on the correlation analysis of multi-sensor signals, and diagnosing whether the state of a train key component is abnormal or not; the vibration signals serve as signals for further diagnosis at the cloud, state features of the vibration signals are extracted at the edge end, online optimization of sparse coding is carried out, and the vibration signals are transmitted to the cloud to serve as diagnosis signal features for further diagnosis; real-time training precision and sample complexity are integrated into online estimation of a model training process, optimal features are transferred from a source domain-edge end to a target domain-cloud end, and the cloud end diagnoses the type and grade of a fault by means of the complex operation capability of the cloud end; a dynamic aggregation strategy is provided, the features of a prediction model are determined, optimization of nodes is enhanced, and a feature network is made to adapt to and complete an online prediction task.

The invention discloses a wind power plant yaw control optimization method based on a sequential optimization algorithm, and the method comprises the steps: obtaining the coordinates of all fans at a current inflow wind direction angle, queuing all fans after the coordinates are converted in sequence, and calculating the yaw angle range of each cycle according to the actual demands; (n-1) times of outer layer circulation are set for the first (n-1) fans in the preliminary optimization process, in each outer layer circulation, yaw angle traversal value taking circulation is conducted on a certain fan according to the obtained yaw range and the uniform step length, and then the optimal yaw combination and the corresponding power of the wind power plant are obtained; and then optimization is carried out to obtain the optimal yaw angle of each fan in the whole field. According to the method, the wake flow secondary deflection effect in the wind power plant and the influence effect of the inflow turbulence intensity on the wake flow are considered, the improved Gaussian wake flow model is adopted to comprehensively consider calculation of wake flow loss in the three-dimensional direction, the calculation result is more accurate, and the final control optimization effect can also conform to the actual engineering situation.

The invention belongs to the technical field of data asset management, and discloses an enterprise data asset auditing method and system based on big data, and the method comprises the steps: collecting financial information data, business operation data, management decision data and external industry data of an enterprise; preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain a financial feature data set, a business operation feature data set, a management decision feature data set and an external industry feature data set; performing feature extraction and weighted fusion on the financial feature data set, the business operation feature data set, the management decision feature data set and the external industry feature data set to obtain a comprehensive feature data set; constructing an audit risk prediction model, inputting the comprehensive feature data set into the audit risk prediction model, and predicting to obtain an audit risk coefficient of the enterprise data assets; and a more comprehensive view angle is provided, and the data asset risk of the enterprise can be evaluated more accurately.

The invention provides a subway pedestrian anomaly detection method and system based on an improved yov8 network and a clip model, and relates to the technical field of subway pedestrian detection, and the method comprises the steps: obtaining a video frame in a subway scene, and carrying out the preprocessing of the video frame; inputting the preprocessed frame data into an improved yolov8 neural network model to obtain an abnormal behavior target detection frame, and performing target candidate tracking on the target detection frame by using a target tracking algorithm to obtain a candidate image of a target area of a specific behavior; the method comprises the following steps: inputting a candidate image containing a target area into a clip model, inputting a set detection text into the clip model, mapping the detection text and the candidate image of the target area, performing text-image similarity matching, and optimizing a similarity matching threshold by adopting a Q-Learning reinforcement learning immune algorithm to obtain a similarity matching result; and the abnormal behavior detection accuracy is improved.

The invention discloses a residual period prediction method, system and device for enhancing time sequence prediction and a medium, and aims to solve the problems of low efficiency and limited prediction precision in the prior art in a long time sequence prediction task. According to the method, the periodic residual error is predicted by explicitly modeling the periodic mode in the time sequence data and combining the prediction model, so that the prediction precision and efficiency of the time sequence prediction model are improved. The method comprises the steps of determining the period length of a data set, generating a learnable cycle period, separating a periodic component and a residual component of a time sequence, performing prediction based on a residual, and finally combining a prediction value of the residual component and the corresponding periodic component to form a final prediction result. The method can effectively capture the inherent periodicity in the time sequence, reduce the prediction error, significantly improve the precision of long-time sequence prediction, maintain the simplicity and high efficiency of the model, and is especially suitable for application scenarios with clear periodicity characteristics, such as power consumption, traffic flow, meteorological data and the like.

The invention discloses a parallel iteration Runge-Kutta method for simulating a second-order nonlinear differential equation. The parallel iteration Runge-Kutta method comprises the steps that a simulated ordinary differential equation is converted into a target nonlinear implicit differential equation, and a system differential algebraic equation is established; judging whether a system matrix and a decomposition iteration matrix in the system differential algebraic equation are updated or not; a four-order extrapolation method is adopted to obtain a predicted differential variable derivative, and if a prediction result is not converged, parallel iterative operation is carried out through a four-stage four-order Runge-Kutta method; after the iteration process is converged, judging whether an operation result accords with a preset error range, if not, updating a system matrix according to a step size reduction strategy and re-decomposing an iteration matrix, and if yes, updating the operation result; and storing and displaying an operation result when judging that simulation is finished. The algorithm stability and the calculation efficiency of the rigidity problem can be considered.

The invention belongs to the field of time sequence anomaly detection, and discloses a time sequence anomaly detection method and system based on frequency enhanced convolution Transform, and the method comprises the steps: carrying out the sequence-level preprocessing of an input sequence, including normalization and block generation; encoding the input sequence by adopting a multi-head self-attention mechanism; extracting interrelationships in the sequence by a full convolution module having a plurality of stacked 1D convolution layers of different kernel sizes; s2 and S3 are repeated, and multi-layer coding is carried out on the sequence; converting the encoded embedding into a frequency signal by using a Fourier transform method; the KL divergence and the absolute error are respectively used for calculating the consistency loss of the time domain and the frequency domain, calculating the point-by-point anomaly score, and identifying whether the point is abnormal or not. Wide experiments on five common data sets show that the method is superior to an existing method in the aspect of abnormal recognition.

The embodiment of the invention provides a driving intention recognition method and device, equipment and a storage medium, and relates to the technical field of intelligent driving. The method comprises the steps of obtaining and preprocessing historical vehicle power data and historical operation data of a real vehicle to obtain standardized data, performing screening processing on the standardized data to obtain target data, establishing a short-term driving data prediction model based on a Mamba module, training the short-term driving data prediction model by using the target data, and obtaining a short-term driving data prediction result. Obtaining a target short-time driving data prediction model; obtaining vehicle power data and operation data of a real vehicle, and inputting the vehicle power data and operation data into the target short-time driving data prediction model for processing to obtain a prediction result; and obtaining preset rules corresponding to the plurality of driving intentions, and identifying the prediction result by using the preset rules to obtain a target driving intention. The method is used for achieving the effects of reducing the calculation cost in the driving intention recognition process and improving the accuracy and reliability of driving intention recognition.

The invention discloses a numerical calculation method for multi-scale residual stress and application, and belongs to the technical field of residual stress analysis. The calculation method comprises the following steps: (1) according to the structure and the manufacturing process of a component sample, establishing a temperature field calculation model, respectively setting corresponding moving heat source models, and calculating the macroscopic temperature field distribution of the whole component through finite element simulation; (2) establishing a multi-physical mechanism macroscopic residual stress calculation model based on temperature-stress-microstructure coupling, and calculating a macroscopic stress field by using an indirect thermal-mechanical coupling method; and (3) establishing a multi-mechanism mesoscopic and microcosmic residual stress calculation model based on temperature-stress-microstructure coupling by utilizing a sub-model technology, and calculating mesoscopic and microcosmic stress fields. According to the method, microstructure evolution characteristics and macroscopic thermal boundary conditions in the component manufacturing process are coupled, a multi-scale residual stress calculation method based on a multi-physical mechanism is established, and the distribution rule of residual stress is researched from multiple scales.

The invention belongs to the field of computer vision, and particularly relates to an unmanned aerial vehicle aerial photography night image enhancement method based on long and short exposure fusion. Comprising the following steps: firstly, constructing a low-illumination synthetic data set; secondly, carrying out data preprocessing on the long-exposure image and the short-exposure image in the low-illumination synthesized data set to obtain a long-exposure image and a short-exposure image containing blurring and noise; then, a lightweight denoising and deblurring network is constructed, the first stage comprises a multi-scale input module, a dual-core feature fusion module and a feature reconstruction module, and the second stage adopts a channel attention mechanism module for fine tuning; designing a multi-scale perception loss function, and guiding network training; and finally, using the trained final model to enhance the to-be-tested unmanned aerial vehicle aerial night image. According to the method, the visual effect and fidelity of the aerial image of the low-illumination unmanned aerial vehicle are effectively improved by utilizing complementary information of continuous short exposure and long exposure.

The invention relates to the technical field of plate shell structures, in particular to a plate shell curve reinforced structure design method and a plate shell curve reinforced structure. According to the method, based on given design parameters, an implicit curved surface and a cross section of the curve reinforced structure are selected, then section parameters of the curve reinforced structure are optimized through layout optimization of the curve reinforced structure and establishment of an agent model, and finally the optimal curve reinforced structure design is obtained. The implicit curved surface in the method can be given by any general curved surface form, and has great degree of freedom. The application agent model is high in calculation efficiency and can explicitly process constraints, and the calculation amount is reduced. According to the design method, the E-AMS method is combined, high-precision reduction of the implicit curve can be achieved while explicit is achieved, and accurate modeling of the curve reinforcement structure is achieved. The plate shell curve reinforced structure provided by the invention is obtained by adopting the plate shell curve reinforced structure design method, the design cost is lower, the efficiency is higher, and the structure has better anti-buckling bearing capacity and bearing characteristic.

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 relates to a method for constructing a space-time multi-task prediction large model based on data fusion. The method comprises the steps that firstly, a data fusion module is constructed, the data fusion module comprises a time sequence alignment sub-module, a region fusion sub-module and a feature splicing sub-module, and original data sets are fused based on the three dimensions of time, space and features; then, constructing an input embedding module, and extracting time features of the data set while embedding the fused data set into a high-dimensional space; constructing a spatial coding module, and extracting spatial features of the data set; then, a base large model prediction module is constructed, and reasoning prediction is carried out; and finally, designing a multi-task output module, and performing multi-task splitting on a prediction result to obtain a space-time multi-task prediction large model. According to the method, multiple tasks can be predicted at the same time, the multi-task prediction capability of the model is improved, the calculation cost and deployment cost are reduced, and the method has great advantages in an urban traffic comprehensive scene needing quick response.

The invention discloses an oral cancer multi-task diagnosis system based on a Raman spectrum, and belongs to the technical field of biomedical detection. Comprising an acquisition module and an oral cancer multi-task diagnosis module, and the acquisition module is used for acquiring Raman spectrum data; the oral cancer multi-task diagnosis module is used for processing the Raman spectrum data through a trained oral cancer multi-task diagnosis model so as to generate and output a tumor staging result, a lymph node staging result and a histological staging result; the oral cancer multi-task diagnosis model comprises a bidirectional Mama branch and a residual attention fusion branch, local feature information in the Raman spectrum data is extracted by using the residual attention fusion branch, and multi-scale features are generated by fusing global feature information extracted by the bidirectional Mama branch. Feature information in Raman spectrum data can be effectively extracted, the accuracy of oral cancer diagnosis is improved, and multi-task diagnosis of oral cancer is realized; the problems that multi-task common diagnosis cannot be realized and the diagnosis accuracy needs to be improved in the prior art are solved.

The invention provides a machine learning potential function design method for UO2 and a fission product thereof, and the method comprises the steps: S1, constructing a UO2-Xe / I unit cell with the fission product, and carrying out the optimization of the UO2-Xe / I unit cell; s2, the optimized UO2-Xe / I unit cell structure is disturbed and simulated, and a training data set is generated; s3, establishing a machine learning potential function; s4, performing molecular dynamics simulation on the machine learning potential function; s5, performing density functional theory calculation on the candidate configuration to obtain accurate structure-energy-force data, and repeating the steps S3 to S5 to perform iterative calculation; s6, verifying the trained machine learning potential function to obtain an accurate machine learning potential function; and S7, performing molecular dynamics simulation by using an accurate machine learning potential function. The precise DP model is obtained through machine learning, and a theoretical basis is provided for research of a diffusion mechanism of fission products in UO2 and molecular dynamics simulation of other mechanical properties.

The invention discloses an unmanned vehicle exploration path planning method based on a multilayer fusion map. The method comprises the following steps: S1, constructing a multi-layer fusion map: constructing a multi-layer map through segmentation and layering of point cloud data, and superposing the multi-layer map to form the multi-layer fusion map; s2, generalized Voronoi diagram extraction: identifying a false free space in the fused map, incorporating the false free space into an obstacle, and constructing a generalized Voronoi diagram by calculating the distance between a point and the obstacle; and S3, path planning based on the Voronoi diagram: performing path planning based on the generalized Voronoi diagram by adopting a fast random exploration tree algorithm. According to the invention, efficient and autonomous exploration of the UGV in an unknown three-dimensional space can be realized. Through accurate map construction and intelligent path planning, the UGV can perform efficient navigation and exploration in an unknown environment, and meanwhile, the exploration area and efficiency are maximized. The method has an important practical value for rescue, exploration, monitoring and other applications in a complex environment.

The embodiment of the invention discloses a three-dimensional reconstruction method and a corresponding device. The three-dimensional reconstruction method comprises the following steps: acquiring target area images of a plurality of view angles; generating dense point clouds by using the target area images of the multiple visual angles; performing Gaussian splashing processing on the dense point cloud to obtain a three-dimensional model of the target area; wherein part of Gaussian kernels are deleted according to the importance degree of each Gaussian kernel in the Gaussian splashing treatment, and the importance degree of the Gaussian kernels is determined at least based on the opacity of the Gaussian kernels. According to the method, part of Gaussian kernels are deleted according to the importance degree of the Gaussian kernels in Gaussian splashing processing, and the operation cost is saved under the condition that the three-dimensional reconstruction quality is guaranteed.

The invention provides a hilly and mountainous area tracked vehicle intelligent chassis system and algorithm based on environmental perception and road surface recognition, and belongs to the technical field of agricultural mechanical equipment. The system comprises a mechanical structure part and a control system part. According to the algorithm, multi-modal sensing and fusion decision making of the mobile terminal are achieved, the FPN + ResNet50 algorithm, the IPM image inverse perspective algorithm, the 3D sparse convolutional graph neural network, the TSDF truncation signature distance algorithm, the MDU algorithm, the gating modal sensing convolutional network and the Pi-Sigma fuzzy neural network learning algorithm are innovatively combined, the recognition precision and the decision making reliability are guaranteed, and meanwhile, the fusion decision making efficiency of the mobile terminal is improved. And the calculation amount and the calculation cost are reduced, and the recognition efficiency is improved, so that more accurate, timely and reliable recognition perception and feedback control are realized. The algorithm comprises the steps of image acquisition, image feature extraction, feature fusion, feature comparison and the like, and the method is not only suitable for field operation of tracked vehicles in complex hilly and mountainous areas, but also suitable for transfer driving of the tracked vehicles in a working environment.