Classification of Hyperspectral Images with CNN in Agricultural Lands

Hyperspectral images (HSI) offer detailed spectral reflectance information about sensed objects under favour of hundreds of narrow spectral bands. HSI have a leading role on a broad range of applications, such as forestry, agriculture, geology and environmental sciences. Monitoring and managing of agricultural lands has a great importance on meeting nutritional and other needs of rapidly and continuously increasing world’s population. In this case, classification of HSI is an effective way to creating land use and land cover maps fast and accurately. In recent years, classifying of HSI with convolutional neural networks (CNN) which is a sub-field of deep learning become a very popular research topic and several CNN architectures were developed by researchers. The aim of this study is to investigate the classification performance of CNN model on agricultural HSI scenes. For this purpose, a 3D-2D CNN framework and well-known support vector machine (SVM) model were compared by using Indian Pines and Salinas Scene datasets that contain crop and mixed vegetation classes. As a result of this study, using of 3D-2D CNN has a superior performance on classifying agricultural HSI datasets.