CHICKEN DISEASE CLASSIFICATION USING DEEP CONVOLUTIONAL NEURAL NETWORKS

Abstract

Our nation's poultry industry is growing, and numerous small enterprises have chosen to enter the chicken farming sector. Different chicken diseases, such as New Castle Disease, Coccidiosis, and Salmonellosis, can affect this sector. Timely and accurate classification remains a major challenge because of the lack of experts and the high material costs for diagnosis. Machine learning techniques have proven to be effective in early poultry disease classification. However, previous studies did not focus on object detection, potentially reducing accuracy owing to non target objects in images that affect training. This study aims to develop a Deep Convolutional Neural Network (CNN) model to diagnose poultry by detecting and classifying images of healthy and unhealthy chicken droppings. It applies both experimental and applied research designs using a mixed research approach. We collected 8625 fecal image data points from different poultry farms in the Gamo and Gurage zones for image labeling agricultural support experts participated. The data were split into 70%, 10%, and 20 ratios for training, validation, and testing, respectively. Six fundamental algorithms were utilized to build the model: five for image classification and one for object detection (including the YOLO-V8 object detection algorithm and Deep CNN, MobilenetV3, XceptionNet, Denenet121, and EifficentNetB3 for image classification models). The Region of Interest (ROI) from fecal images was segmented using YOLO-V8, and the resulting segmented image was then categorized into four health conditions (healthy, Coccidiosis, Salmonella, and New Castle disease) using a Deep CNN and pre-trained models. According to the results, YOLO-V8 achieved 93% object identification accuracy. The pre-trained EfficientNetB3 framework achieved 97.2% accuracy by surpassing other pre trained, and base models trained from scratch, demonstrating an optimal performance and potential value for classifying chicken diseases. This study will aid the chicken production industry by offering an automated tool for rapid and precise disease diagnosis