Seafood quality inspection is critical for ensuring food safety and minimizing economic losses from spoilage. While traditional methods are slow and labor-intensive, computer vision and machine learning have emerged as efficient automated alternatives. This study presents SFFDNet, a software engineering-driven convolutional neural network featuring a lightweight 19-layer architecture with optimized feature extraction blocks and regularization strategies. With only 2.49 million parameters—significantly fewer than VGG16 (138M) and ResNet50 (25.6M)—our model achieves 98.80% accuracy on the Large-Scale Fish Segmentation and Classification Dataset. SFFDNet outperforms both transfer learning m... More >