Crop Land Change Monitoring Based on Deep Learning Algorithm Using Multi-temporal Hyperspectral Images

Change detection is done with the purpose of analyzing two or more images of a region that has been obtained at different times which is Generally one of the most important applications of satellite imagery is urban development, environmental inspection, agricultural monitoring, hazard assessment, and natural disaster. The purpose of using deep learning algorithms, in particular, convolutional neural networks and hyperspectral imagery, here is to identify the planting area because these networks have an excellent performance in achieving change detection. In this research, we investigate to use of deep learning methods in comparison with another tradition methods for obtaining changes in an agricultural area so that, after generating difference images with the use of Otsu algorithm we generate a preliminary binary map. Then we extracted the feature by using sparse auto encoder networks and classified pixels in two categories to change and no change by using the convolutional neural networks too. In the end, we obtain a final change map by making a model and evaluation of accuracy. That we have achieved even better results, which indicates the need to use deep learning methods. Since solving, the problem manually related to change detection. To investigate capable of the proposed method, 2 datasets hyperspectral imagery from the American Hermiston agricultural fields in the United States was used and vegetation cover near the Shadegan wetland located in the south of Khuzestan province, evaluated by the Hyperion sensor. The proposed method compared to other methods has an overall accuracy of 95% and the kappa coefficient of 0.86.