Considering the vast area of the watershed area for soil and water conservation, the disasters such as sloping land collapse and earth-rock sliding are produced in large areas of bare land due to typhoon, heavy rain or illegal development. It often directly or indirectly influences the water quality of the water source or cause serious
reservoir siltation which threaten the usage for water resources of people. Presently, the management of the catchment area requires manpower to regularly inspect for illegal use, which is time-consuming and labor-intensive, and also costs a lot of money. Therefore, this study is combined with image recognition technology to conduct land monitoring through image recognition technology, which greatly saves manpower expenditure. The image of the study area in this study is in the Wulai Mountains. This area has a series of monthly image data, and two months are taken for analysis and research (April ~ May 2019).The band adopts the 4 basic bands, and the landform adopts 7 classification categories such as trees, turf, and buildings.In the first step, we used the Support Vector Machine (SVM) in machine learning. The Convolutional Neural Networks(CNN) is a deep learning approach for the second step of data training. This study adopts the Alex-Net model of CNN to carried out the calculation progress. The model uses 5 layers of convolution layers and 3 layers of pooling layers as the overall architecture. The differences between CNN and SVM are drawn by confusion matrix and thematic map. As shown by the research results, the classification performance of the CNN is in the same level as the SVM for some landforms (95%). However, the integrity is good and there is no serious noise with classification error, especially considering the two topographical parts of the buildings and bare-lands.
Key Words: Machine learning, Deep learning, Confusion matrix