XAI based Enhanced Fish Disease Detection using Multi-Feature Extraction in Varied Color Spaces

Fish aquaculture is essential in ensuring global food security, but it encounters significant challenges due to variety of fish diseases. There has not been much progress made in creating automated mechanisms for identifying fish diseases and resolving related issues. Early disease detection in fish is essential for preventing the disease from spreading and infecting other organisms. This study focuses on developing a reliable fish disease detection model by combining some feature extraction methods such as Local Binary Patterns (LBP), Block-wise Truncation (BWT), Hough Transform (HT) and Local Directional Patterns (LDP). Features were extracted from different color spaces like RGB, HSV, LAB and YCbCr to enhance the accuracy and robustness of the work and study the effect of varying color spaces on the disease detection strategy. Features were then fed to various machine learning classifiers such as SVM, Random Forest, Gradient Boosting, AdaBoost, Decision Tree and Logistic Regression and multiple metrics such as Accuracy, recall, F1-Score, and precision were used to evaluate the performance of these algorithms. Two datasets containing images of diseased fish were used to assess the performance of the models. In the first dataset Random Forest classifier achieved highest accuracy of 100% in YCbCr and LAB color space whereas in second dataset Ada Boost classifiers achieved highest accuracy of 91.43 % in YCbCr color space. The findings of this study show that utilizing these several feature extraction techniques in combination with classification algorithms significantly improves classification accuracy, enabling a dependable and effective strategy for automated fish disease diagnosis. Feature analysis showed that BWT features comprised the most important feature in our classification model, with LBP features coming in second. This accounts for explainable AI approach of the ML models used.