Identify Animal lumpy Skin Disease Using Image Processing and Machine Learning

Abstract

Ethiopia has the largest livestock population in Africa. However, productivity of the sector in Ethiopia has multifaceted constraints; the Lumpy Skin Disease is one of the major factors. Lumpy Skin Disease is known as a major risk to cattle production and substantial impacts on livelihoods and food security especially for our country. Currently, detection of Lumpy Skin Disease in our country is assessed manually. However, manual evaluation takes significant amount of time and requires trained professional and experienced person. Therefore, technology is needed to prevent animal disease epidemics. Automated detection of Animal Lumpy Skin Disease has advantages over the manual technique. Detection of Lumpy Skin Disease in Cows is developed in literature. But Animal Lumpy skin disease has different classification based on its severity. There is a need to further identify the different stages of Lumpy skin disease to know to what extent the animal is affected by lumpy skin disease. In this study, Lumpy skin disease detection model is constructed using Convolutional Neural Network (CNN) for feature extraction and SVM for classification. CNN is the state of the art for deep feature extraction, hence we used it for feature extraction. The model used to detect and classify animal Lumpy Skin Disease skin diseases into Severe, Mild and Normal. The dataset is collected from Oromia region Bale zone Medawelabu wereda and Arsi zone Chole wereda Livestock production offices and from internet external images repository. After collecting data, Image augmentation, Image Preprocessing, and Image Segmentation techniques are applied to enhance image quality and identify region of interest. During image preprocessing, the image is resized to 200x200. Gaussian filtering is applied to remove noise and Histogram equalization to balance the intensity of image. Adaptive thresholding segmentation method is used to identify region of interest. Out of the total 1740 image dataset, 80% is used for training and 20% for testing. Experimental results show that, SVM classifier outperforms RF(Random Forest) and Softmax classifiers. Quantitatively, an overall accuracy of 95.7% is achieved by using SVM classifier; on the other hand, RF achieves 87. 4% and Softmax classifier achieves accuracy 94.8%. Noises in the image is a challenging task for properly detecting the region of interest and hence we recommend as a way forward to use advanced noise removal techniques to improve image quality for proper segmentation and Lumpy skin disease detection.