Multilabel land cover aerial image classification using convolutional neural networks

Classifying the remote sensing images requires a deeper understanding of imagery, machine learning classification algorithms, and profound insight into satellite images’ know-how properties. In this paper, convolutional neural network (CNN) is designed to classify multispectral SAT-4 four classes: trees, grassland, barren land, others. an airborne dataset that captures in 4 bands (R, G, B, infrared). The proposed CNN classifier learns image’s spectral spatial properties from ground truth samples provided. contribution paper three-fold. (1) A framework for feature extraction normalization built. (2) Nine different architectures models are built, multiple experiments conducted images. (3) image structure resolution captured by varying optimizers CNN. correlation between classes identified. experimental study shows vegetation health predicted most accurately models. It significantly differentiates grassland tree vegetation, which better than other classical methods. tabulated results show state-of-the-art analysis done learn land cover