A Deep Learning-Based Framework for the Detection of Big-4 Snakes

Snakebite incidents pose a significant public health concern, with venomous snake species contributing to fatalities and medical emergencies. India accounts for the highest number of snakebites and is called the capital of snakebites globally. Despite advancements in medical treatment and antivenom supplies, precise and timely identification of snake species remains difficult. Among venomous snake species, 'Big-4' snakes known as spectacled cobra, Russell's viper, common krait, and saw-scaled viper, stand out due to their prevalence and lethal potential. Computer-aided diagnosis has recently proven helpful in analyzing snake images and in the field of herpetology. This research article evaluates the performance of deep learning models, including VGG16, DenseNet, InceptionV3, and EfficientNet, in accurately classifying the Big4 snake species using image data. The evaluation of the proposed models is conducted using 1000 snake images collected from iNaturalist.org and the Kaggle repository. In the proposed work, EfficientNet has achieved an accuracy rate of 97.69%, surpassing the performance of VGG16, DenseNet, and InceptionV3 models. The EfficientNet demonstrated substantial computational efficiency, with each training epoch requiring approximately 43 seconds. In comparison, DenseNet demonstrated an accuracy rate of 93.69%, followed by InceptionV3 with 90.92%. Moreover, VGG16 attained an accuracy rate of 90.04%. These results highlight the potential of EfficientNet as a robust solution for accurate snake species classification, offering both high accuracy and computational efficiency. Overall, the proposed study contributes valuable insights into utilizing deep learning techniques to address critical challenges in snakebite management and wildlife conservation.