The influence of CLAHE on the accuracy stability of the automatic classification of Mars surface lineament structure based on DEM image
Ziyi Li, Pengcheng Yan, Jiarui Liang, Xiaolin Tian · Journal of Physics Conference Series · 2021
Abstract This article introduces training ResNet to automatically classify the lineament structure of Mars surface (DEM image), and then use CLAHE to pre-process the samples to improve the stability of accuracy. The linear structure of Mars surface is mainly divided into two types: dorsum and vallis, and crater is added as a representative of non-lineament structure. We have prepared a sample set of 300 samples for each class, divided into training set, test set, and validation set at 6:2:2. Without pre-processing, the highest accuracy rate reached 98.33% (for crater), 100.00% (for dorsum), 98.33% (for vallis), 89.44% (for total). However, the accuracy of Dorsum fluctuates greatly and frequently. After CLAHE pre-processing, the fluctuation of the accuracy of dorsum is significantly reduced.