433 results about "Model complexity" patented technology

The invention provides a multi-modal sentiment analysis method based on knowledge distillation and a dynamic fusion mechanism. The multi-modal sentiment analysis method comprises the following steps: pre-training a single-modal teacher model; a single-mode teacher model is used for guiding the learning of a multi-mode student model, and through an interactive knowledge distillation mechanism, the middle layer probability distribution of the teacher model is used as a target to learn the correlation between modes and the cross-mode characteristics; interactive knowledge distillation comprises three loss functions: firstly, calculating the difference of output distribution of output layers of a teacher model and a student model, and defining the difference as cross-modal knowledge distillation loss; secondly, adding alignment loss based on a real label, and constraining a prediction result of the student model to be close to a real emotion label in a cross entropy form; and finally, introducing label smoothing loss to soften the real label. According to the method, the pre-trained single-mode teacher model with relatively good performance is stored and is used for guiding the learning of the multi-mode student model, meanwhile, the loss function is introduced to optimize the multi-mode student model, so that the multi-mode student model is gradually aligned with the output distribution of the teacher model, and meanwhile, the adaptive capacity of the multi-mode student model to the modal heterogeneity is enhanced. The single-mode teacher model greatly reduces the complexity of the model, reduces the calculation amount, and has better performance in the field of multi-mode sentiment analysis.

The invention discloses a strategy question-answering system and method based on a vertical domain large model, and the method comprises the steps: S1, receiving the natural language input of a user, and retrieving an open source data set to obtain an original data set with the comprehensive similarity meeting the requirements; s2, performing necessary data cleaning and preprocessing on the original data set to obtain a training set; s3, performing fine tuning on the pre-processed basic large language model by adopting the training data set, and introducing a low-rank structure to modify a weight matrix to obtain a final model; s4, the problem enters a final model for post-processing to generate prediction output; the system takes a large language basic model subjected to fine tuning training as a core, introduces a low-rank structure to perform fine tuning on model parameters, reduces parameter quantity required by training while keeping model performance, reduces model complexity, reduces an overfitting risk, improves generalization ability, supports deep fusion of a specific field knowledge base and universal field data, and has a wide application prospect. And the accuracy and correlation of questions and answers are improved.

The invention discloses a modulation and signal category identification method based on multi-scale attention and residual error, and relates to the technical field of signal type and modulation mode identification, and the method comprises the steps: obtaining signal data sets under different signal types and modulation modes, and dividing the signal data sets into a training set and a verification set; constructing an end-to-end deep learning model based on an input preprocessing module, a shared convolutional feature extraction module, a multi-task branch module and a joint loss optimization module; the shared convolution feature extraction module comprises a channel expansion convolution layer, a multi-scale attention residual module and a down-sampling module; an end-to-end deep learning model is trained; and inputting the data of the to-be-tested signal into the model to obtain the signal type and the modulation mode of the to-be-tested signal, and through end-to-end deep learning model processing, the problems of model redundancy, computing resource waste, insufficient inter-task information utilization and the like can be solved, and the model complexity and the training overhead are reduced while the identification accuracy is improved.

The invention discloses an industrial product surface defect detection method based on feature coupling, and the method comprises the steps: 1) constructing a four-stage backbone feature extraction network, and integrating a DPSC module to expand a receptive field, thereby achieving the gradual feature learning from local to global; 2) designing a multi-scale feature fusion network, realizing effective fusion of different scale features, and performing deep coupling on high-level semantic information and low-level detail information; 3) constructing a multi-branch detection head to realize full-scale coverage; and 4) performing end-to-end training optimization: calculating a weight importance score by applying an LAMP pruning strategy, deleting redundant parameters, and remarkably reducing the model complexity and the calculation cost while keeping the detection performance. According to the method, through a global-local feature coupling mechanism, the technical problems that in industrial product surface defect detection, the defects are highly similar to the background, and the scale change is large are effectively solved, and high-precision and high-efficiency defect detection is achieved.

The invention discloses an efficient target detection method in a foggy environment, and relates to the technical field of target detection. According to the method, a lightweight foggy day target detection model is established and called DF-DETR, a defogging module of a double-branch structure is designed through an edge enhancement module, and target features of a foggy day image are effectively captured; then, a dual convolution feature extraction module DualConv-Block is designed, so that feature extraction is enhanced, and meanwhile, the complexity and the calculation amount of the model are remarkably reduced; besides, an EAA attention mechanism is combined with an intra-scale feature interaction module to form an AIFI-EAA module, and the AIFI-EAA module is integrated into the hybrid encoder, so that the attention capability of the model on dense targets is improved, and missing detection and false detection are effectively reduced; finally, a dynamic sampling scale attention feature fusion module is designed, alignment of multi-scale features is achieved through dynamic up-sampling, the flexibility and robustness of feature expression are enhanced, and the fusion and expression ability of the multi-scale features is further optimized.

The invention relates to the technical field of cognitive impairment detection, and discloses a cognitive impairment early warning method and device based on an electroencephalogram micro state and an eye movement trajectory, and the method comprises the following steps: S1, data acquisition, S2, electroencephalogram preprocessing, S3, electroencephalogram feature extraction, S4, eye movement feature extraction, S5, feature fusion, and S6, early warning judgment. According to the method, through independent convolution branch of electroencephalogram and eye movement features, a receptive field is expanded by utilizing cavity convolution to capture multi-scale features, and long-distance dependence is modeled by a self-attention layer; after tensor splicing, time sequence information is dynamically fused through a gating cycle unit (GRU), and cross-modal time correlation is captured. The method has the advantages that the multi-modal feature hierarchical extraction and self-adaptive modeling capability is enhanced, the complementarity fusion efficiency is optimized, meanwhile, by means of cavity convolution sparse connection, self-attention parameter sharing and GRU lightweight design, the model complexity and the calculation efficiency are balanced, and efficient feature representation is provided for cognitive impairment early warning.

The invention discloses a wind power prediction system for optimizing a neural network based on a genetic algorithm, relates to the technical field of new energy power system prediction, and improves the precision and adaptability of wind power prediction by fusing the genetic algorithm and a deep neural network. The system adopts multi-objective genetic optimization, randomly initializes a neural network parameter combination, evaluates the fitness by taking a prediction error and model complexity as double objectives, and screens out an optimal network architecture through evolution operation; in the aspect of neural network training, the system adopts an LSTM and TCN hybrid network as a basic model, dynamic weighting input features of a meteorological attention mechanism are combined, a learning rate and regularization parameters are optimized by using a genetic algorithm, model convergence is accelerated, and overfitting is prevented; in addition, for the space-time imbalance of the wind power data, a generative adversarial network is introduced to generate synthetic data in an extreme weather scene, and the generalization ability of the model is enhanced.

The invention discloses a light-weight YOLOv8n model partial discharge type detection and classification method based on a PRPD spectrogram, and aims to solve the problems of low partial discharge type detection precision, high model complexity, large calculation amount, poor real-time performance and the like in the prior art. According to the method, PRPD spectrograms of four typical partial discharge types of tip, air gap, suspension and surface are collected, normalization, gray processing, data enhancement and expansion and other preprocessing are carried out, a LabelImg platform is utilized to carry out labeling, then a YOLOv8n model is subjected to lightweight improvement, ShuffleNet-V2 is adopted as a trunk feature extraction network, a CoordAttention mechanism is introduced to enhance the feature extraction capability, and the feature extraction efficiency is improved. An EIOU loss function is applied to optimize target frame regression precision; the partial discharge type detection and classification precision is improved, the model calculation complexity and parameter quantity are reduced, real-time detection is realized, and the method has important practical application value.

The invention discloses an SOP defect detection method based on lightweight multi-scale feature fusion, and the method specifically comprises the steps: obtaining an SOP defect image, and marking the defect image; a YOLO-DBS model is provided on the basis of the YOLOv8n; replacing a C2f module in a backbone network with a DCFB module, designing an efficient multi-scale feature fusion pyramid network BiFPN, and adding an SENetV2 module in an input layer of a detection head; training the improved algorithm model by adopting an SOP packaging chip surface defect image training set; and transmitting a defect image test set into the trained model, recording a detection result and evaluating the performance of the model. According to the method, the balance between the precision and the model complexity is realized while the high detection precision is maintained.

The invention discloses a power adaptive digital pre-distortion method, system and device based on sparse GRU and a medium. The method comprises the steps that input and output signals under different powers are collected, a composite feature vector is constructed through amplitude normalization and phase alignment, and a pre-distortion training target is generated in combination with indirect learning; constructing a sparse GRU neural network, introducing L1 regularization constraint, and obtaining a sparse pre-distortion model through joint optimization of mean square error and sparse constraint; partial weight updating is realized based on power grading, and weight fusion output is carried out during power interval switching so as to complete power adaptive optimization; pruning and compressing the sparse GRU model and then deploying the sparse GRU model in a pre-distortion module to compensate the nonlinearity of the power amplifier; according to the method, the model complexity is reduced through rarefaction, the dynamic working condition stability is improved through a power adaptive mechanism, pruning compression adapts to low-power-consumption hardware, high precision, low complexity and adaptive capacity are considered, and the method is suitable for the low-power-consumption and high-performance requirements of a broadband wireless communication system.

The invention relates to the technical field of remote sensing image processing, in particular to a multi-modal target detection method and device based on attention self-modulation fusion, and the method comprises the steps: extracting visible light modal shallow layer features of a visible light modal image, and extracting infrared modal shallow layer features of an infrared modal image; carrying out fusion processing on the visible light modal shallow layer features and the infrared modal shallow layer features to obtain target fusion features, and carrying out feature extraction to obtain deep semantic features; and performing multi-scale feature aggregation on the deep semantic feature, performing feature enhancement on a feature aggregation result based on a preset feature enhancement mechanism to obtain a multi-scale fusion feature, and detecting the multi-scale fusion feature to obtain a detection result. Therefore, the problems of high calculation cost and high model complexity of multi-modal target detection in the background technology are solved, and efficient and lightweight multi-modal target detection is realized.

The invention discloses a water supply system scheduling optimization method and device, electronic equipment and a storage medium, and relates to the field of intelligent scheduling of water supply systems. The method comprises the steps of collecting and preprocessing multi-dimensional data to obtain a data set; the pressure data is analyzed from the space-time dimension, and the unfavorable points and the pressure requirements thereof are accurately identified; the method comprises the following steps: constructing a water volume prediction model by adopting a time sequence model, and constructing a total water head difference prediction model by adopting a Light GBM gradient boosting tree in combination with a MultiOutputRegressor multi-output regression framework; constructing a minimum total water production cost objective function based on a prediction result, and outputting an optimal scheduling scheme by combining water volume and pressure constraint iterative optimization; and establishing a model updating mechanism to ensure dynamic adaptation of the strategy. According to the method, the problems of insufficient pressure guarantee, extensive cost control and weak model practicability and generalization ability are solved, the inherent contradiction that a traditional mechanism model is high in complexity and a pure data driving pressure prediction model is poor in generalization and lacks physical significance is overcome, and safe, stable and efficient intelligent technical support is provided for a water supply system.

The invention provides an artificial intelligence model service method and system based on edge computing, and relates to the technical field of artificial intelligence. Performing feature analysis on the original data and dividing into subsets; selecting a matched target model for each subset; determining a priority according to the model complexity and the computing resources; executing a model generation result according to the priority; processing results are summarized and returned to the mobile terminal. According to the method, intelligent processing of different feature data by the edge computing node is realized, the computing efficiency and the resource utilization rate are improved, and the flexibility and adaptability of artificial intelligence service are enhanced.

The invention discloses a turbine spindle bearing fault diagnosis method, and belongs to the technical field of turbine fault diagnosis. The method solves the problems that the fault diagnosis accuracy of an existing method is low, and an existing model is difficult to deploy and apply in an actual industrial scene due to a complex structure. According to the noise adaptive random convolution block, the model is guided to form inductive bias for global information through random disturbance on local features of vibration signals, the sensitivity to noise is remarkably reduced while key features of a fault are reserved, and meanwhile, the complexity of the model is simplified. The global attention mechanism can effectively enhance the attention capability of the model on key fault features, and has good noise identification capability. Meanwhile, the model has good nonlinear dynamic characteristic modeling capability, high-precision turbine main shaft bearing fault identification can still be realized in a high-noise environment, and the number of model parameters is small. The method can be applied to turbine fault diagnosis.

The invention discloses a deep counterfeit video detection method fusing ViT and spatial features, belongs to the technical field of video identification, and is used for detecting a deep counterfeit video. The method comprises the following steps: constructing a neural network fusing ViT and spatial features; training a deep counterfeit video detection network; acquiring face data information in the deep forged video; and inputting the processed video information into the trained deep forged video detection network, and outputting whether the video information belongs to a forged video or not. According to the method, the advantages of the convolutional neural network in image counterfeiting detail extraction in video counterfeiting detection are fully utilized, thoughts of orthogonal convolution, an attention mechanism, residual connection and the like are combined, the accuracy of deep counterfeiting video detection is improved while the complexity of the model is maintained, and the detection efficiency of the deep counterfeiting video is improved. And moreover, the method has relatively stable detection performance in videos generated by various counterfeiting technologies.

The invention is suitable for the technical field of computer vision and deep learning, and provides a multi-scale feature fusion concrete defect detection method based on improved SAM, and the method comprises the steps: obtaining a concrete defect image data set; preprocessing the image data set and dividing the image data set into a training set, a verification set and a test set; the YOLOv9 is trained to automatically detect concrete defects; constructing an improved SAM model, replacing a prompt encoder of the SAM with YOLOv9 and edge detection, and modifying a mask encoder of the SAM; the training effect of the model is evaluated through the four indexes of the accuracy rate, the recall rate, the F1 value and the intersection-to-union ratio, and the improved SAM model is optimized according to the training effect. By improving the structure of the SAM, the detection capability of the model on small defects can be enhanced, and the detection accuracy and efficiency are improved. In addition, according to the method, the complexity of the model can be reduced, and the training efficiency and generalization ability of the model are improved.

The invention discloses a distributed photovoltaic power prediction method and system, belongs to the technical field of renewable energy prediction, and solves the prediction precision and efficiency bottlenecks of a traditional model under complex meteorological conditions through fusion of deep feature learning and an adaptive optimization mechanism. Firstly, a time sequence sample set is constructed based on a sliding window mechanism, and non-stationary fluctuation characteristics of a power sequence are dynamically captured; eliminating the dimensional difference between the input features and the tags through minimum-maximum normalization; constructing a four-level restricted Boltzmann machine stacking structure, and extracting time-space coupling characteristics of power data by layer-by-layer unsupervised pre-training; and designing an improved extreme learning machine dynamic analysis architecture, optimizing the number of neurons in a hidden layer in combination with grid search, and realizing global optimal balance between model complexity and prediction precision. According to the method, an effective dynamic mode and noise interference are distinguished by using hierarchical feature abstraction capability of the RBM, prediction robustness in strong fluctuation scenes such as cloudy and rainy scenes is remarkably improved through analytical solution and parameter adaptive adjustment of the ELM, and efficient technical support is provided for intelligent scheduling and energy storage optimization in a high-proportion photovoltaic grid-connected background.

The invention discloses a virtual space multi-person interaction synchronous control method oriented to an end-cloud collaborative architecture. The invention relates to a computer graphics and three-dimensional modeling technology, and discloses a cross-format lightweight and geometric consistency maintenance method based on a three-dimensional model. Through format-independent geometric representation and a self-adaptive lightweight strategy, efficient compression and precision maintenance of three-dimensional model cross-format conversion are realized. The method specifically comprises the following steps: performing format analysis and geometric feature extraction on an input model, and establishing a unified internal representation; adaptively selecting a multi-level LOD lightweight strategy based on the complexity of the model; the accuracy of key information is ensured through geometric feature keeping and topology consistency detection; the geometric consistency is dynamically maintained by combining error monitoring and an iterative correction mechanism; and generating a target format lightweight model and carrying out quality verification. According to the method, adaptive precision control, multi-level consistency maintenance and format irrelevant processing are combined, the model size and conversion errors are effectively reduced, and cross-platform compatibility and geometric fidelity are improved. The method can be widely applied to the fields of industrial design, game development, virtual reality and the like.

The invention relates to a lightweight double-flow cross-modal interaction RGB-D saliency target detection method, which comprises the following steps: S1, data preparation: obtaining an RGB-D data set of a task for training and testing, taking a part of an NJU2K data set and a part of an NLPR data set as a training set, and taking the training set as a training set; taking the rest of the NJU2K data set, the rest of the NLPR data set, the rest of the SIP data set, the rest of the STERE data set and the rest of the SSD data set as test sets; s2, constructing a network model: S21, constructing a feature extraction backbone network of a decoder, S22, constructing a self-adaptive cross-modal fusion module (ACM), and S23, constructing a multi-scale cavity attention module (MSA); the method comprises the following steps: S24, setting a decoder, S25, calculating a loss function, and S26, evaluating indexes; the method has the advantages that high calculation efficiency can still be kept under the condition of low model complexity, and meanwhile high generalization and accuracy are achieved when multiple types of scenes are processed.

The invention discloses an apple leaf disease segmentation method based on a lightweight dual-path network, an apple leaf disease segmentation device based on the lightweight dual-path network, an apple leaf disease segmentation device and a computer readable storage medium. The problems that an existing disease segmentation method is high in model complexity and insufficient in multi-scale recognition capability are effectively solved. The lightweight encoder adopts a depth separable convolution and channel recombination technology, so that the parameter quantity and the calculation complexity are greatly reduced while the feature extraction capability is maintained, and the model can be deployed on edge equipment such as an unmanned aerial vehicle and a field robot. The enhanced cavity space pyramid pooling module constructs abundant multi-scale receptive fields through multi-branch parallel cavity convolution with different expansion rates, and can capture feature information of initial tiny disease spots and later fused disease spots at the same time.

The invention provides a flight path prediction method fusing space perception and time-frequency conversion, and relates to the technical field of computers and air traffic management. The method comprises the following steps: firstly, constructing a flight path data set, preprocessing a complete flight path in the flight path data set to obtain a flight path sequence, and generating a training sample; a flight path prediction model is constructed for flight path prediction; the method comprises the following steps: firstly, carrying out feature dimension expansion on a flight path sequence by a flight path prediction model to obtain flight path high-dimensional feature representation; and learning a spatial structure dependency relationship and a time dependency relationship in the flight path sequence high-dimensional feature representation to generate a final flight path prediction result. And finally, constructing a loss function of the flight path prediction model based on the idea of supervised learning, and training the constructed flight path prediction model to obtain a trained flight path prediction model. According to the method, the prediction accuracy is improved while the model complexity is reduced.

The invention discloses a distributed anti-interference fault-tolerant control method and a distributed anti-interference fault-tolerant control system for a multi-six-rotor unmanned aerial vehicle attitude system. The control method comprises the following steps: constructing a multi-six-rotor unmanned aerial vehicle attitude model based on a multi-six-rotor unmanned aerial vehicle attitude system model and interference uncertainty; designing a disturbance observer which is used for estimating the degree of external disturbance and actuator faults; and based on the output of the multi-six-rotor unmanned aerial vehicle attitude model and the degree of external interference and actuator fault, constructing a Lyapunov function by using a backstepping recursion technology, and designing a distributed anti-interference fault-tolerant controller for generating a control output signal to the multi-six-rotor unmanned aerial vehicle attitude model. According to the method, the Lyapunov function is constructed, so that output signals of all followers and leaders are synchronous, and errors are converged. Meanwhile, the general approximation characteristic of the neural network is utilized, the nonlinear term and the unmodeled dynamic are approximated online, and the model complexity is reduced.

The invention relates to an EDSR (Enhanced Depth Super-Resolution) image super-resolution reconstruction method based on particle swarm optimization, and the method comprises the following steps of: (1) carrying out super-resolution reconstruction on an EDSR (Enhanced Depth Super-Resolution) image; the method comprises the following steps: firstly, inputting a low-resolution image data set as a training sample, defining an optimization space containing the number and stage of residual blocks, convolutional layer parameters, an attention module, an up-sampling mode and the like, initializing particle swarm optimization (PSO) parameters, and dynamically constructing a candidate network by particle position coding; a candidate network is dynamically constructed through particle position coding, and a residual block layer, a convolution layer, an attention module and an up-sampling module are sequentially configured. The candidate network is subjected to limited training, and individual and global optimal positions are updated through fitness function evaluation fusing PSNR and model complexity. And finally, a global optimal structure is selected for complete training for low-resolution image reconstruction, the detail reduction capability and the reasoning efficiency are remarkably improved, and a high-quality image is generated.

The invention discloses a multi-target prediction method, system and device for a financial time sequence and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: firstly obtaining financial time sequence data and auxiliary information, and carrying out the vector conversion and fusion to form a unified input representation; extracting features through a shared encoder, and generating a global feature vector; and inputting the data to a multi-task prediction output head to realize parallel prediction of the prediction tasks. According to the scheme, cross-task feature sharing is realized by using the shared encoder, the modeling collaboration is improved, and resource waste and prediction conflicts are avoided; through data fusion, the depiction capability of complex financial dynamics is enhanced; the generalization and robustness of the model are improved by adopting a lightweight output head, the method is suitable for diversified financial prediction scenes, and the prediction efficiency and accuracy are improved while the model complexity is reduced.

The invention discloses a lightweight identification method for hole diseases in a road, and the method comprises the steps: constructing a road B-Scan radar image data set which accords with a YOLO data set standard, and dividing the data set into a training set, a verification set and a test set; based on an improved YOLO11 target detection network, a training set is used for model training, a backbone network, a neck network and a head network are optimized, and a PConv module, an ADown module and an EMA attention mechanism are introduced to improve the calculation efficiency and precision. According to the method, the recognition accuracy of the road internal cavity disease image is remarkably improved, and meanwhile, the model complexity and the calculation overhead are reduced. The trained target identification model can efficiently process the internal cavity disease image of the to-be-detected road, and an accurate identification result is provided. Experiments show that the processing speed is greatly improved while high detection precision is kept, and an effective solution is provided for intelligent detection of the holes in the road.

The invention discloses a light-weight small target detection method and system for images shot by an unmanned aerial vehicle, and the method specifically comprises the steps: constructing a neural network architecture which comprises a backbone network, a feature aggregation network and a detection head; improvement of light weight and attention enhancement is implemented in the backbone network, and a multi-scale initial feature map is extracted; constructing a feature aggregation network Neck, performing cross-level fusion and refining processing on the multi-scale initial feature map, and outputting a refined feature map; a lightweight target detection head Head is constructed in combination with a large-kernel depth separable convolution module and a special decoupling head structure of a YOLOv11 network, and decoupling prediction is performed on the refined feature map; training is carried out by adopting a mixed loss function based on a normalized Wasserstein distance and modulated IoU, a trained lightweight network is obtained, and detection of a lightweight small target is realized. According to the invention, the complexity of the model is reduced, the detection speed is improved, the high detection precision is maintained, and the method is suitable for real-time detection tasks on an unmanned aerial vehicle resource limited platform.

The invention discloses a multi-view heterogeneous cascade non-stationary time sequence prediction method based on Mama improvement, and belongs to the technical field of time sequence analysis. The prediction method comprises the following steps: collecting and preprocessing time sequence data of a target domain; the time sequence data are stabilized and decomposed; the decomposed seasonal part is embedded from a univariate view angle and a multivariate view angle respectively; an embedding result is correspondingly input into a Mama encoder and a multi-granularity cascade Mama encoder decoder heterogeneous module for feature learning; performing stationarity correction on the features based on an autocorrelation matrix; and predicting the feature representation after stability correction and the decomposed trend part, adding prediction results, and carrying out inverse normalization to obtain a final prediction result. The method provided by the invention solves the technical problem that the trend and periodicity of dynamic evolution in data are difficult to capture when an existing method faces a non-stationary time sequence, and also solves the problems that an existing model is high in complexity and low in prediction accuracy.

The invention provides a bearing life prediction method, and belongs to the technical field of equipment life prediction, and the method comprises the steps: obtaining a bearing health state index, and comprehensively calculating a comprehensive health score CHS of a certain time step; based on the comprehensive health score CHS, constructing a prediction model of the bearing health state to the remaining use time; training the constructed prediction model by using a comprehensive health score (CHS); and the trained prediction model is used to identify the dynamic characteristics of the bearing health state to the remaining use time, and the bearing health state is predicted according to the trend of historical health state scoring so as to complete the prediction of the bearing life. According to the method, the problems of high model complexity, insufficient explanation, poor applicability and universality and large over-fitting risk in the prior art are solved, so that high precision, high robustness and high reliability of bearing life prediction are realized.

The invention provides an inland ship target detection method and system, and relates to the technical field of ship image recognition. The method comprises the following steps: establishing an inland ship target data set; performing data enhancement processing on the inland ship target data set; a BiFPN module is adopted to replace a PANet module in the YOLOv8 model; a SimAm attention module is introduced into the neck network; the detection head is improved by adopting RepConv convolution; the Shape-IoU is introduced to replace the CIoU to serve as a bounding box loss calculation function of the YOLOv8 model; and after the improved YOLOv8 model is obtained, an inland ship target detection model is obtained after inland ship target data set training, and target detection of the to-be-detected image is realized. The method can effectively balance the detection precision and the model complexity under the condition of ensuring the real-time performance, and has obvious advantages in small target detection and complex environments.

The invention discloses an embedded AI intelligent computing power architecture method, and particularly relates to the technical field of artificial intelligence processing architecture. Collecting a resource state parameter set R of the embedded device; obtaining a to-be-executed AI task set T, wherein each task comprises model complexity, real-time requirements, expected response duration and priority; constructing a computing power resource allocation evaluation function F based on R and T, and outputting a task scheduling priority score; allocating tasks to the embedded AI module according to the allocation scheme and executing reasoning; the resource state is dynamically monitored in the task running process, and if it is predicted that resources are about to be overloaded, a scheduling function F is triggered to reconstruct a resource allocation scheme; performing iterative optimization on weight parameters in the function F based on task history feedback; by means of the method and device, optimal adaptation of multi-task concurrent scheduling can be achieved under the condition that resources are limited, the computing power resource utilization rate, the response efficiency and the system stability are improved, and the method and device are suitable for various edge side AI scenes.