Vision-based texture and color analysis of waterbody images using computer vision and deep learning techniques

Abstract Vision-based analysis of waterbodies can provide important information required for monitoring, analyzing, and managing water resource systems, such as visual flood detection, delineation, mapping. Water, however, is an ornery object in image processing, it be found different forms colors nature. This makes the classification, tracking images videos difficult computer vision models. There are still differences resulting from texture its inherent optical properties associated with which recognized extracted to support models better analyze images. study aims utilize a set early, mid-level, high-level techniques, including Gabor kernels, local binary patterns (LBPs), deep learning (DL) extract color digital For this purpose, ATLANTIS TeXture (ATeX), dataset classification analysis, was used. Models were trained task on ATeX. Then, performance each model extracting features evaluated compared. Results showed that accuracy achieved by magnitude tensor, LBP, DL (ShuffleNet V2 × 1.0) 29, 35, 92%, respectively, thus outperforms traditional vision-based techniques. Moreover, results raw represented spaces (e.g., RGB, HSV, etc.) emphasized importance processing water. Analyzing representative types facilitate designing customized Convolutional Neural Networks (CNNs) scenes, CNNs recognize objects through both shape clues their relationship entire field view.