Detecting the Simultaneous Occurrence of Strawberry Fungal Leaf Diseases with a Deep Normalized CNN

Deperias Kerre, Henry Muchiri

Abstract: Crop diseases are a major threat to food security and economic development, but their rapid identification remains difficult. Strawberry is one of the cash crops grown in Kenya and has a substantial contribution to the country's income with 50% of the foreign earnings from the horticultural crops being attributed to this crop. However, strawberry farming in Kenya has been adversely affected by the prevalence of strawberry fungal leaf diseases mainly leaf Blight and leaf Scorch which occur and affect the strawberry leaves simultaneously. A review of existing computer vision models that have been leveraged for the detection of these diseases discovered that none of the models has the capability to detect leaf Blight and leaf Scorch diseases especially when they occur simultaneously on the same leaf. This makes their detection a challenge. In this paper, we present a deep Convolutional Neural Network (CNN) model for detecting the simultaneous occurrence of Strawberry Leaf Spot and Leaf Blight. The model presents a novel technique of detecting more than a single class of the strawberry fungal diseases on the same leaf. A dataset containing a total of 1,134 images was used in training and evaluating the model. The model achieved an accuracy of 98%, precision of 98.9%, a recall of 93.3% and an f1-score of 95.9% overall thus demonstrating the feasibility of this approach. The performance of the CNN model was also compared with that of other machine learning algorithms which include Support Vector Machine(SVM), K-Nearest Neighbor(KNN) and the Random Forest. The comparison also included the other existing CNN architectures which are GoogleNet, Resnet and VGG.