The Role of Deep Learning and Hybrid Approaches in Food Quality Control: Automatic Classification of Fish Freshness
Tarık Talan, Habibe Aslan · Sakarya University Journal of Computer and Information Sciences · 2026
Not cached yet. This record was fetched live for this request and is stored after the lookup. Reopening the page should serve the cached copy.
Sakarya University Journal of Computer and Information Sciences · OpenAlex · article · 2026
Abstract
Regular monitoring of fish freshness is critical to product reliability, consumer health, and compliance with food industry quality standards and quality control processes. In this study, fish were classified as fresh or non-fresh using image-based methods. To determine the model that best predicted fish freshness, eight models were compared: four deep learning (DL) architectures, three hybrid models, and one ensemble model. The dataset consisted of images of each fish taken from approximately 20 different angles. Data leakage, which emerged during initial tests, enabled the models to distinguish freshness classes more easily than expected.
To address this issue, a group-based splitting method was applied. All models in the study were evaluated with a single, common training protocol under the same regularization, epoch, and optimization settings. Dropout, early stopping, and image enhancement (rotation, color jittering, and horizontal flipping) were used during training. According to the results, the ensemble model performed best with 98.44% accuracy. This was followed by ResNet50 (97.10%) and MobileNetV2 (94.31%) models. Grad-CAM and Layer-CAM analyses revealed that the models focused on regions that indicate freshness, such as the fish’s eyes, gills, and surface texture.
Open the resolved locator · 10.35377/saucis...1839329
Metadata licence: OpenAlex metadata (CC0)