Universitas Dian Nuswantoro conducted a study on determining an augmentation method for a Convolutional Neural Network (CNN) using the FOX algorithm for fish image classification. The activity involved UDINUS researchers Wahyudi, Firman; Soeleman, Moch Arief; Pramunendar, Ricardus Anggi; and Andono, Pulung Nurtantio as participants. The study addressed the automatic identification of fish species from digital images. This identification is relevant to conservation biology, fisheries management, and biological research.
The activity considered factors that affect the achievement of high classification accuracy. These factors included the amount and quality of data, image processing methods, feature extraction techniques, classification algorithms, and optimization methods. UDINUS proposed a CNN model optimized using the FOX algorithm to select the best augmentation method. The selected augmentation techniques were Kmeans Color Quantization, Horizontal Flip, Voronoi, Elastic Transformation, and Contrast Normalization.
The results showed fish species recognition accuracy of up to 98.75% during training. The proposed model also achieved a validation accuracy of 96.90%. The source states that this result indicated minimal overfitting and strong generalization capability. The approach required intensive training time but was described as highly effective for applications demanding high accuracy and good generalization. The source further states that the approach enhanced understanding and management of marine ecosystems in support of sustainable fishing practices. This activity therefore presents UDINUS research on fish image classification, augmentation method selection, and the use of FOX optimization in a CNN model.