Akademik Veri Yönetim Sistemi
5th International Conference on Research of Agricultural and Food Technologies (I-CRAFT’2025, GANJA/AZERBAYCAN) , Baku, Azerbaycan, 20 - 24 Ekim 2025, ss.38, (Özet Bildiri)
Fish and seafood are an integral part of the world's diet since they are rich in proteins, polyunsaturated fatty acids, vitamins, and minerals. Their perishable state, however, leaves them extremely vulnerable to physical, microbiological, and metabolic degradation while being handled and stored. Therefore, maintaining freshness and quality is a top priority in the food, aquaculture, and fishing industries. This study aims to predict the freshness of sea bass (Dicentrarchus labrax) fillets using deep learning models based on image data. For this purpose, 10 fillets were monitored daily from the day of purchase until the third day after spoilage. Each fillet was imaged from six different angles, and corresponding color and sensory analyses were performed. In total, classification models were developed for 7 categorical parameters. Three pre-trained transfer learning models (VGG19, ResNet50, and EfficientNetB0) were employed. For each parameter, all image angles and transfer models were evaluated separately, resulting in the development of 18 prediction models. The findings revealed that the VGG19 model achieved the best overall performance. Ultimately, a model capable of predicting the spoilage status of fillets (purchasability parameter) with an accuracy of 83% was obtained. This research is financed by the Scientific Research Projects Unit of Cukurova University (FBA-2024-16557).