Radar Echo Rainfall Intensity Classification Algorithm Based On Architecture Searc

Weather radar is an important means of rainfall monitoring,and accurate rainfall intensity classification based on weather radar echo characteristics is crucial.However,traditional rainfall weather recognition methods mainly rely on image data from weather radar,extracting image features through algorithms such as image binarization and threshold segmentation,and then relying on subjective human recognition,which is time-consuming and prone to errors.In order to overcome these limitations,a neural architecture search method based on 3D attention module is proposed to classify rainfall radar echoes.The genetic algorithm is mainly used as the optimizer to search for the optimal architecture,and the optimal convolutional neural network structure is accurately found through the rapid optimization of the width and depth of convolutional neural network architecture,so as to realize the classification and identification of rainfall intensity,accurately extract the rainfall characteristics of each grade,effectively identify the rainfall grade,and help relevant people to obtain the rainfall classification results from radar echoes.In this algorithm,two basic blocks are designed:convolutional block and pooling block,which are encoded as chromosomes.Genetic algorithm is used to search for the optimal individual through a series of cross mutations.In order to effectively suppress the edge interference pixel problem in radar echo rainfall classification and optimize the convolutional output,the algorithm introduces a 3D attention module to improve the perception ability of the model.In order to focus on features of different levels and reduce the risk of overfitting,multiscale convolution is introduced to extract features of different scales,reduce the amount of computation and improve the efficiency of computation.In order to suppress invalid information and focus on effective information,confidence attention is also introduced to enhance the model’s attention to key areas and improve the classification ability of the model.Based on these,this paper proposes a neural architecture search algorithm based on multi-scale convolution and confidence attention mechanism.In this algorithm,three basic blocks are designed: convolutional block,pooling block and confidence attention block.Through flexible coding and decoding schemes,the optimal network architecture is searched to improve the accuracy of rainfall radar classification.The algorithm uses automated techniques to find the best convolutional neural network structure for classifying rainfall intensity.Compared with manually designed convolutional neural networks,this algorithm can achieve better performance.Through a large number of experiments on the two data sets,it is proved that the proposed two algorithms have better performance than the traditional comparison algorithm.