Image Processing and Supervised Learning for Efficient Detection of Animal Diseases

ABSTRACT  
Animal disease management such as trypanosomiasis can be a frustrating proposition if proper techniques are not employed for the disease management as lack of information on diseases that can affect individual animals or an entire herd is tantamount to economic loss. Classifiable symptoms associated with the disease help strategize for problem identification and solution provision. Currently, disease management, symptoms classification and diagnoses are manually performed and are time consuming due to the fact that it takes longer time for infected animal to be manually diagnosed especially those animals which are remote from veterinary. These challenges motivate the development of detection tools that can perform automatically using deep learning approaches such as convolutional neural networks which have received great acceptance in literature but are computationally expensive and consumes huge amount of training data. This paper seeks to improve on the deep learning methods of managing animal diseases by using non-deep methods such as random forests and support vector machines with adoption of pre-processing and thorough testing methodology. Test carried out on 1000 images shows random forests system performing better with achievement of 95.20% accuracy. The system’s accuracy, recall, and precision are higher than the previous non-deep approaches and performing convolutional neural networks approach. To our best knowledge, this work is one of the newest works carried out to facilitate the detection of animal diseases for the benefit of animal husbandry; this implies that more research efforts are ongoing to improve this area of research for reliable policies and practices. Unavailability of huge dataset is the greatest challenge of the research which limited the extent to which comparison of the proposed method to others could have been made possible.