Paper Title
A COMPARATIVE ANALYSIS OF DEEP CNN ARCHITECTURES FOR AUTOMATED FISH DISEASE DIAGNOSIS IN AQUACULTURE

Abstract
The wellbeing of fish populations in the water is important in the wellbeing of the ecological balance of the water and the viability of aquaculture in the water. The most important thing in the management and mitigation strategies of fish diseases is the timely and accurate detection of the disease. This article introduces a new method of detecting fish diseases using the transfer learning and ensemble models. In order to construct the correct model, we have collected a total of 4742 images representing the eight classes. The paper thoroughly assesses five pre-trained CNN models, including InceptionV3, InceptionResNetV2, Modified MobileNetV2, DenseNet121 and NASNet Mobile to multi-classify fish diseases. The method of multiclass image processing, normalization, feature extraction with ML to identify fish diseases is shown to be by far more effective in terms of sensitivity and specificity. Of all the five models that displayed competent performance, Modified MobileNetV2 was most dependable with a peak of 97% accuracy. It did better than any other method that happened to be tested, proved to be highly efficient in detecting diseases in fish and in making precise disease diagnoses. Our results highlight the opportunity of deep learning in improving fish disease detection systems, which promises opportunities in future studies on aquatic health management and on aquaculture. Keywords - Aquaculture, Deep Learning, InceptionV3, Modified MobileNetV2.