Coral Health Assessment Using Modified Inception V3 and DenseNet 121

This study compares the training performance of modified Inception V3and DenseN et121 models for coral health assessment. DenseNet121 exhibits a gradual improvement in accuracy and a significant reduction in validation loss towards the end of training. Despite these improvements, the model faces considerable fluctuations in validation loss and accuracy early in the process, with a peak validation accuracy of88.37<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. In contrast, the modified Inception V3model demonstrates a more stable and consistently high performance throughout its training. It achieves a peak validation accuracy of 99.14%, with validation loss steadily decreasing and maintaining minimal fluctuation. This stability suggests that InceptionV3 converges more quickly and effectively. Overall, modified InceptionV3 outperforms DenseNet121 in terms of peak validation accuracy and consistency, proving to be a more robust and reliable choice for coral reef health assessment tasks.