YOLO-R50: A Robust Approach for Wheat Ears Detection and Bacterial Black Chaff Classification

The detection and identification of bacterial black chaff disease in wheat ears is very important for crop field management. The previous studies use the machine learning and deep learning methods to recognize the wheat ears based on color, shape and texture features. This paper focuses on developing and testing a robust approach for wheat ear object recognition and classification, with a specific emphasis on distinguishing between healthy wheat ears and those infected with Bacterial Black Chaff (BBC) disease. Using the YOLO-R–50 (You Only Look Once with ResNet-50) model, this study examines the effectiveness of single YOLO, YOLOv7, and YOLO-R50 architectures in detecting and localizing wheat BBC disease in agricultural fields. We evaluate the efficacy of these models and give a comparative study of their precision values in properly detecting and localizing BBC disease-infected wheat ears. The proposed methodology entails training the YOLO-R50 model using a dataset of annotated wheat field photos, which include both healthy wheat ears and those affected with BBC disease. Following model training, extensive evaluation methods are carried out to determine the models’ ability to reliably detect and categorize wheat ears, with a particular emphasis on detecting the presence of BBC disease. In the case of there is more annotated wheat images, the performance is also increased. The proposed YOLO-R50 achieves the rate of high accuracy 98.59% and rate of sensitivity 97.81%, segmentation rate 96.40% and specificity 91.89% when compared to existing single YOLO, and YOLO version 7.