Enhanced Classification of Very High-Resolution Remote Sensing Image Using Ensemble Specific Class Convolutional Neural Networks

The classification of very high-resolution remote sensing images is complex because of their intricate and detailed characteristics, necessitating exact classification techniques. The proposed method employs a class-specific ensemble (ESC) technique to classify very high-resolution (VHR) remote sensing images. The proposed method was compared to standard convolutional neural network (SC) classification methods, bagging with majority voting (BV) and weighted bagging (BW). The experiment was conducted 30 times to collect a comprehensive dataset to verify the proposed method’s reliability and precision. The ESC method was assessed using the paired t-test against the SC, BV, and BW methods under identical trial conditions. The results demonstrate that the ESC method outperforms the SC and BV methods at a significance level of 0.05 and the BW method at a significance level of $\mathbf{0. 1 0}$. The $\mathbf{t}$-test outcomes verified the presence of a substantial disparity in mean accuracy between the ESC and the other approaches. The results revealed the benefits of using the ESC method as an alternative, more practical method for classifying VHR images, resulting in enhanced performance in remote-sensing applications.