Research And Implementation Of Remote Sensing Image Multi-Land Cover Types Recognition Technology Based On Deep Learning

Remote sensing generally refers to the technique of observing the ground from artificial satellites or flight equipment,perceiving the main characteristics of the target,and analyzing them through the propagation and reception of electromagnetic waves(including light waves).In practical applications,remote sensing technology is widely used in many aspects such as resource survey,surface environment monitoring,and human activity monitoring because of its advantages of real-time and low cost.However,the identification of land types from remote sensing images still faces difficulties so far.Firstly,due to the vast area of China,the same land type in different regions may present different image characteristics,and the problem of same-spectrum and same-spectrum dissimilarity arises when parsing remote sensing images;in addition,multiple land types are interspersed and the boundary distribution is disordered,which makes the image parsing results have voids and fine patches.The above problems limit the recognition accuracy of existing methods.To address the above problems,this paper proposes a deep learning-based remote sensing recognition method for multiple land types,and the main work is as follows:Firstly,this paper proposes a convolutional neural network(CNN)model called Dual Path Attention Network(DPA-Net).CNN)model.This model has a simple modular structure and can be added to any conventional deep learning segmentation model to enhance its ability to learn features.Specifically,in this paper,two types of self-attentive mechanism modules are attached to the segmentation model,one focusing on spatial information and the other on channel information.Later,the outputs of these two attention modules are fused to further improve the network's ability to extract features,which helps to obtain more accurate segmentation results.Second,data preprocessing and enhancement strategies are used to compensate for the small number and uneven distribution of datasets.The proposed network in this paper is tested on a high-resolution image dataset(GID).The experimental results show that the trained recognition results of the optimized model in this paper are all better than those of the pre-optimized model.Therefore,it is proved that the semantic segmentation model based on the self-attention mechanism can be applied to the land type recognition of remote sensing images with little increase in computational effort compared with the common semantic segmentation model.In addition,this paper implements a prototype system for multi-land type recognition of remote sensing images using the improved method,which is convenient for remote sensing practitioners to study and use.