Analyzing the Impact of Image Size on CNN Model Performance for Fish Disease Diagnosis

The rapid growth of aquaculture has made the diagnosis of fish diseases a critical challenge, potentially leading to significant economic losses. This study investigates the impact of image size on the performance of Convolutional Neural Networks (CNNs) for fish disease diagnosis, with the specific goal of determining the optimal image size for maximizing CNN performance. While CNNs have shown excellent performance in image classification tasks, their effectiveness can vary significantly depending on factors such as image size. This study analyzes how image downscaling affects CNN performance by evaluating predictive accuracy across various image sizes. The experimental results confirm that reducing image size increases predictive accuracy and reduces variance in model performance. Specifically, images scaled down to 20×25 pixels achieved an average accuracy of 90.18%, surpassing the original image accuracy of 83.17%. Smaller images help reduce unnecessary noise, prevent overfitting, and operate effectively in environments with limited computational resources. The findings underscore the necessity of selecting optimal image sizes to maximize diagnostic performance and practicality. This research provides valuable insights into designing more efficient and accurate automated diagnostic systems in aquaculture and highlights the importance of carefully considering input image size when developing CNN models. The results contribute to advancing machine learning applications in aquaculture, offering effective management and mitigation strategies for fish disease outbreaks.