Multispectral Remote Sensing Image Classification Using Modern Machine Intelligence Approach

Classification of multispectral images is a processing tool used in various applications of remote sensing in natural hazards assessment, agricultural and environmental monitoring to assess and analyze current events so as to avoid damage caused by untimely disasters in nature. Several image classifiers have been presented in the past, some of which are supervised classifiers and some unsupervised ones that have been applied to different remote sensing images to classify them. But the classification accuracy on any of the multispectral images is not good enough to identify the real and correct image classes. Therefore, the real and true classification of multispectral images remains a challenging task even today. To accomplish this challenging task, we have presented a novel machine intelligence approach in this paper. The proposed approach is based on the pixels of multispectral images in which the spectral properties of the image have not been changed. The model consists of a total of five layers consisting of three hidden layers in which seven features vectors are applied to the input layer and all hidden layers are activated with the ReLU function. There are four neurons on the output layer, which is activated with the softmax function. The accuracy of the proposed model was achieved at 99.97% with a 0.9997 Kappa coefficient which is 9.04% higher than the previously proposed models. Due to the good accuracy of this model, it can be used for the accurate assessment of any natural disaster and agricultural health monitoring.