Enhanced Segmentation Accuracy in High-Resolution Remote Sensing Images Using a Multi-Scale Convolutional Network

Abstract

Due to the traditional high-resolution Remote Sensing Image Segmentation (RSIS), the efficiency of the model algorithm is very low and the accuracy is also very poor. We design a new machine learning segmentation network architecture composed of a full convolutional network framework. The whole structure is tightly connected, each layer can feed back to each other and the multi-scale convolution kernel is used to build a wider network to improve the adaptability of the network at different scales. Compared with other traditional models, it has higher segmentation accuracy. This paper also optimizes and improves the algorithm used in the model, which makes the algorithm in this paper have more excellent accuracy and recall and is superior to the traditional algorithm in all aspects and has more outstanding performance. The experimental results show that compared with the traditional models and algorithms, the accuracy of the proposed model for high-resolution RSIS is up to about 95% and it has good stability and less interference from the outside world. It is superior to the traditional machine learning segmentation network model in many aspects.