Smartphone-assisted Apple Scab Identification and Quantification Using Artificial Intelligence

Apple scab, caused by the fungus Venturia inaequalis, is the most common fungal disease affecting both wild and cultivated apple species in Pennsylvania. under conducive conditions, losses caused by scab disease can approach 100%. Early identification of apple scabs will allow growers to respond when the symptom first appears, reducing the economic loss from apple damage and the chemical input costs. Plant pathologists use the severity of scab disease, which is the percentage of the scab lesion area divided by the total infected leaf area, to track epidemics, anticipate outbreaks, and evaluate yield losses. In this study, we developed a region-based deep learning model trained on in-field apple scab mobile images for automatic disease segmentation at pixel levels to aid real-time decision-making. During the image acquisition process, the mobile camera lenses were set to parallel the infected leaves to reduce errors caused by image distortion. The trained model was deployed on a web-based application using the Flask application programming interface (API). Disease information in apple orchards can be connected to growers' smartphones via internet browsers. The developed model achieved a mean average precision (mAP) of 70%, with the infected leaf class segmentation mAP of 85%. The results show that a deep learning-based computer vision system can detect apple scab features and infected areas accurately. This system will pave the path for growers toward smartphone-based solutions to identify and quantify apple scabs early and quickly, allowing for a more effective integrated pest management (IPM) strategy in apple orchards.