An Efficient Deep Learning-based InceptionV3 Model for Coral Reef Classification

In this study, the coral classification has been performed which has shown a groundbreaking use of the InceptionV3 model to detect coral health in underwater images. The results show an outstanding accuracy level of more than 97.3%. Utilizing the unique advantages of deep learning, the model has been implemented at different values of learning rates as 0.01, 0.03, 0.05, and 0.07. With the learning rate value of 0.01, the highest accuracy has been achieved, whereas, the least results have been identified with the accuracy of 0.07 learning rate value. The least testing loss value is a mere 0.0253 at epoch 50, demonstrating how powerful and effective the proposed approach is for healthy and bleached coral classification is. The proposed research includes a comprehensive image analysis of various conditions and phases. Its fine-tuned InceptionV3 model demonstrates remarkable efficiency at spotting the hidden patterns revealing the state of coral health. The collected image dataset has been found very small for the deep learning model to get trained, hence each image has been implemented with the two different augmentation operations. With the augmented images dataset, the InceptionV3 has shown its breakthrough accuracy of 97.3% to become the best-performing model for coral reef classification. It is an effective and early detection model, particularly in cases of healthy and bleaching corals. In addition to providing new approaches to predicting coral ecology, this research also brings out the transformative power of deep learning in applications of other domain practices. These results open the door to the incorporation of such models into other underwater species classification applications in which coral health assessments can be performed with greater accuracy.