Canine Skin-Disease Detection Using CNN

Skin diseases in dogs often lead to itching, discomfort, inflammation hair loss etc. So, early detection aids pet owners for early identification and treatment leading to the well-being of pets as well as owners since many diseases can be transmitted from dogs to humans. Since CNN is commonly used for image classification, most of the working models use ResNet, MobileNetV2, VGG-16, and DenseNet in CNN for best results. Existing models usually provide predictions for individual diseases or classify them into larger groups such as bacterial, fungal, or hypersensitivity. These models cannot pinpoint the exact disease of their category (i.e., bacterial, fungal or hypersensitivity). To overcome this limitation, a custom CNN model has been developed which is engineered to achieve optimal accuracy. The proposed model facilitates the detection of the five most prevalent canine diseases, specifically pyoderma, Leprosy, Flea allergy, Earmites, and Ringworm. The proposed model has undergone training on a dataset containing 5000 images and has been evaluated using images from external sources for testing which generated an accuracy of $89.7 \%$.