In Bangladesh, the crab industry is a major economic driver, producing approximately 20,000 metric tons annually with a value of around $100 million. Over 500,000 fishermen and coastal workers depend on crabs, particularly blue swimmers and mud crabs, for their livelihoods. Beyond its economic significance, crab meat also provides substantial nutritional value. Accurate classification of edible and economically valuable crab species is crucial, as proper identification can enhance profitability. Despite existing efforts, current crab detection methods have shown limited success. In this study, we employed deep transfer learning models, specifically Convolutional Neural Networks (CNNs), to classify six types of crabs using a dataset of 1,560 images (1,200 for training, 180 for validation, and 180 for testing). We fine-tuned hyperparameters, including the number of layers, loss function, activation function, learning rate, and epochs. Among the CNN frameworks tested including MobileNet-v2, Xception, Inception-v3, VGG-19, and DenseNet-169, the customized ResNet-50 model, termed ResCrabNet, achieved the highest performance with a training accuracy of 99.20% and validation accuracy of 98.12%. Grad-CAM and LIME techniques were used to visualize the classification results, providing insights into the model’s decision-making process.