Water Classification Using Convolutional Neural Network

Abstract

The classification of water sources is a challenging task due to the low contrast texture features, the visual similarities between them, and the causes posed by image acquisition with different camera angles and placements. The various image enhancement techniques, i.e., Unsharp Masking (UM), Histogram Equalization (HE), Contrast Limited Adaptive Histogram Equalization (CLAHE), and Contrast Stretching, were used to highlight the contrast and texture features of water images. The enhanced image samples were then fed to the proposed Convolutional Neural Network (CNN)-based model named WaterNet (WNet) for classification. From all employed image enhancement techniques, Contrast Limited Adaptive Histogram Equalization (CLAHE) provides better results in terms of contrast and texture features of water. CLAHE also improved the classification performance of the proposed model, with an accuracy of 97%. For comparison, experiments have also been performed on state-of-the-art pre-trained models, which are DenseNet-201, Inception_ResNet_v2, Inception_v3, and Mobile-Net. Comparison shows that the proposed technique achieves better accuracy in comparison with the state-of-the-art methods.