Detection of Powdery Mildew Pest in Apple Tree Leaves Using Deep Learning in Intelligent Sprayer Robots

Abstract

Early detection of plant pests and diseases is crucial in agriculture. Deep learning can be utilized to more accurately diagnose diseases and pests on leaves and other parts of the crop. This study aims to use deep learning and processing of leaf images to identify powdery mildew in apple trees. It proposes a novel method using a transfer learning algorithm that uses minimal input images in 2 classes of powdery mildew and healthy to identify this type of pest with high accuracy. This method is evaluated by applying it to three various convolutional neural network (CNN) architectures, including VGG16 [1], AlexNet [2], and GoogLeNet [3]. The results show that the error resulting from the effect of angle and intensity of illumination in real field images can be effectively removed by this method and delineate that the AlexNet exhibits a higher accuracy of 99.53%.