An Improved Object CNN Method for Classification of High-Resolution Remote Sensing Imagery

Abstract

Land cover and land use (LULC) classification of very fine spatial resolution remote sensing images is a challenging task. Though the object-based convolutional neural network (OCNN) has proven to be an effective method for LULC classification, it still has some shortcomings. For example, it is difficult to achieve a well-segmented result with few parameter adjustments. Besides, the traditional convolutional neural network (CNN) is hard to make full use of the spectral information in the remote sensing images. To this end, we propose an improved LULC classification method. Specifically, an adaptive segmentation algorithm is used to automatically adjust segmentation parameters to achieve the best-segmented results. Due to the different areas and shapes in segmented units, a unique sample extraction method is proposed to better extract representative samples. For better classification, a new CNN model is also constructed to fully use spectral information. The proposed method has been validated on two real remote sensing images and achieved excellent classification performance.