An Automated Mucilage Detection Model Using Deep Convolutional Neural Network: TuncerNeXt

Abstract

Convolutional Neural Networks (CNNs) are distinguished for their exceptional performance in image classification. A number of these models have been developed, drawing inspiration from seminal works. This research introduces an innovative CNN model that integrates attention mechanisms specifically tailored for detecting mucilage on the ocean surface. To facilitate this research, a comprehensive dataset was assembled from 15 disparate ports, segmented into three distinct categories: the presence of mucilage, sea surface without waves, and sea waves. The rationale for including the sea wave category is to augment the accuracy of the proposed CNN model by accounting for the morphological similarities between sea waves and mucilage. The developed model, termed TuncerNeXt, comprises four principal components: a stem, TuncerNeXt blocks, downsampling stages, and an output phase. The novelty of TuncerNeXt resides in its fusion of attention mechanisms with residual blocks, taking cues from the structural design of ConvNeXt's principal block. This innovative approach has resulted in TuncerNeXt being a streamlined CNN model, boasting approximately 2.1 million adjustable parameters, rendering it an efficacious approach for image classification endeavors. Upon evaluation with the compiled dataset, TuncerNeXt achieved a validation accuracy of 97.60% and a test accuracy of 98.66%.