Development of AI-based Detection and Classification model for Apple Crop Diseases

In India, the valley of Kashmir holds the maximum share of Apple production with more than 75% of total apple production. Currently, around 160000 hectares of land in the Valley is under apple cultivation with an annual productivity of around 180000 MTs [source: Directorate of Horticulture, 2021], in which most part is exported to various regions of the world. Apple orchards are under constant threat from various types of viral pathogens, fungus, bacteria, and insects. They continuously damage the apple fruits and leaves, this is primary cause of low apple yield and results in a huge economic loss to the apple industry every year. Diseases like apple Scab, Apple Cedar Rust, Powdery Mildew, apple Blotch and apple Rot remain a major threat for the apple growers. Early diagnosis of apple diseases can help in controlling of infection spread and ensure higher yield, thus preventing substantial economic losses. Therefore, timely detection of diseases is crucial for enhancing both quality and quantity of apples. Traditional manual disease identification and inspection is laborious, time-consuming, error prone and requires a thorough knowledge of apple plant pathogens. Instead, automated approaches save both time and effort. In this research, , developed an AI based detection and classification model for apple crop diseases using deep learning architecture based on transfer learning convolutional neural network called YOLOv5. The model is improved to optimize for both detection speed and accuracy and applied to multi class apple plant disease detection in the real environment. The dataset corpus is formed which consists of 7909 images belonging to 6 classes namely apple scab, apple cedar rust, powdery mildew, healthy, apple blotch and apple rot. Then, the data annotation process is performed on all the images as per their target classes and saves the labels in a text file. During annotation of images in the training dataset, each text file contains information about the target class and the corresponding bounding coordinate. Several data augmentation techniques are performed to enrich and diversify the dataset which improves the model generalizability and eliminates the problem of overfitting. The non-maximum suppression (NMS) algorithm is used with darknet 53 framework. The mean average precision (mAP) and F1-score of the trained detection model is 90.22% and 92.5%, respectively. The developed model can be employed as an effective and efficient method to detect different apple plant diseases under complex orchard scenarios and can be extended to different fruit crops and automated agricultural detection processes.