Application of Intelligent Ground Object Recognition and Classification Method Based on AI Recognition to Foreign Source Risk Assessment in Transmission Line Channels
Jieyin Nan, Yanhong Liang, Zhenhui Chen, Qingyue Huo, Songbo Wang, Xiao He, Shun Zhao · 2023
This paper designs a foreign source classification detection method for transmission line channels based on AI recognition. Firstly, the image is collected, and then the remote sensing image data is preprocessed. The sample set of typical ground feature elements required for training is segmented on remote sensing image, and all the training samples are enhanced. It provides sufficient training data for the subsequent ground object training recognition of network model. Thirdly, different training parameters are selected according to the characteristics of different ground objects. In this algorithm, train_crop_size is adjusted, and a convolutional neural network based on shallow neural network is designed. The parameters are adjusted in terms of cavity convolution rate, down sampling, up sampling, decoding and encoding fusion image. Experiments on real environment samples show that the model can obtain high recognition accuracy and fast recognition rate in multiple spatial scales. It has strong generalization ability of natural scene and practical value of engineering.