A Scene Detection and Classification Model for Remote Sensing Images Using Deep Learning Technique with Water Flow Optimization

Abstract

Scene classification using deep learning (DL) is a common and effective way in RS and geospatial analysis. It is most vital in environmental monitoring, mapping, land planning, and land management. Nevertheless, the current techniques are issues as vulnerability to noise interference, lower classification accuracy, and poor generalization skills. Remote sensing images are frequently used in the description of urban and rural regions, change detection, and other fields. In general, the RSI is high-resolution and covers extensive and diverse data, appropriate analysis of RSIs is most significant. DL systems like Convolutional Neural Networks (CNNs) are exposed significant result in image detection tasks, making them suitable for scene classification in RSIs. So, this study develops a new Water Flow Optimizer with Deep Learning Enabled Scene Detection and Classification (WFODL-SDC) algorithm on RSIs. The main focus of the WFODL-SDC system lies in the optimal detection and classification of various scenes that exist in it. To accomplish this, the WFODL-SDC technique involves an adaptive median filtering (AMF) method for removing the noise that exists in it. Besides, the WFODL-SDC technique uses SE-DenseNet system for the derivation of useful feature vectors. The experimental values inferred that the WFODL-SDC methodology obtains optimal results with other recent approaches. At last, auto encoder (AE) has been executed for the recognition and classification of various kinds of scenes. The simulation result analysis of the WFODL-SDC technique undergoes utilizing benchmark image database.