In the last few years, computer vision has become significantly more adept at identifying image features. However, it is still challenging to classify fish automatically based on similarities between different types and varying factors like their position or lighting. We propose an efficient fish disease detection method based on our training and evaluation results of four other models, including one Deep Hybrid convolutional neural network (CCN) architecture that uses transfer learning techniques for Rohu fish disease detection. Our proposed Deep Hybrid Network is a combination of three transfer learning based CNN models (VGG16, Xception, DenseNet201) that is trained and tested using images of Rohu fish native to Bangladesh. We have collected images of Rohu fish diseases related to 4 different conditions. HYBRID-CNN achieved 99.82 percent accuracy on our dataset.