Detection of Poultry Diseases Using Convolutional Neural Networks: A Fecal Image-Based Approach

Early disease detection is critical for maintaining a healthy poultry flock and ensuring the overall success of poultry production. Traditional methods for disease diagnosis often rely on physical examinations or laboratory tests, which can be time-consuming, expensive, and require specialized expertise. This study presents the generation and evaluation of a Convolutional Neural Network (CNN) model for detecting chicken diseases through fecal images. The generated model was trained using 6,949 labeled images with 128 x 128 dimensions for 500 epochs. The highest training accuracy, 99.73%, was achieved at epoch 418, with a corresponding loss value of 0.0099. In contrast, the validation accuracy reached 98.69%, with a slightly higher loss value of 0.0767. The model demonstrated robust performance on the test set with an overall accuracy of 98.04%. While the classification performance was strong across most categories, some misclassifications occurred, particularly between Healthy and Salmonella samples, likely due to visual similarities between the two classes. The evaluation results for the performance of the model emphasize its effectiveness in identifying poultry diseases with high precision (98.11%), recall (98.04%), F1-score (98.05%), and specificity (100%). Additionally, the model was integrated into a user-friendly application that allows real-time disease diagnosis based on chicken fecal images. This tool holds significant potential for improving disease management and facilitating timely interventions in poultry farming.