Faecal Image-Based Chicken Disease Classification Using Deep Learning Techniques

Abstract

Various diagnostic methods are used in chickens, including counting oocytes in the stool or intestinal tract, detecting the virus and using polymerase chain reaction (PCR) procedures, which require multiple diagnoses. The diseases are transmitted through contaminated feed and excrement from infected poultry. As a result of late diagnoses or a lack of credible specialists, many domesticated birds are lost by farmers. The most common ailments affecting chickens can be easily identified in the pictures of chicken droppings using artificial intelligence and machine learning methods based on computer screening and image analysis. In this paper, a model for early detection and classification of poultry diseases with high accuracy using wildlife database is proposed. The dataset contains 6812 images of four different classes such as healthy chicken, Coccidiosis, Salmonella and Newcastle images is proposed. A deep learning method based on convolutional neural networks (CNN) is used to predict whether chicken faecal image belongs to any of the four categories. The pre-trained DenseNet model, the Inception model and the MobileNet model were used to predict whether chicken faecal belonged to four categories with minimum loss. In comparison with the above-mentioned models, DenseNet method produced the best results with an accuracy of 97% which is recommended for poultry diagnostic application.