Early Disease Detection of Plants Using Deep Learning

Abstract: Due to the estimated yearly loss of 20-40% of the total crop production, plant diseases are a serious threat to the world's food security. Early agronomic interventions are enabled by precise and timely identification of the diseases. This paper presents an end-to-end deep learning model for the early diagnosis and classification of plant leaf diseases. In the suggested model, the publicly available benchmark dataset, "PlantVillage," which contains more than 54,000 leaf images with 17 types of diseases affecting tomato, potato, and corn crops, is used for the development of the suggested model based on a specially designed three-block CNN model, namely "PlantCNN." In the suggested model, the accuracy of the results is improved by the development of a high accuracy classifier based on the refined EfficientNet-B0 model. In the suggested model, the "Random Horizontal Flipping" technique is used for the improvement of the robustness of the model. In the suggested model, the "Disease Severity Index" with Low, Medium, and High levels is calculated based on the HSV color thresholding method for the estimation of the visually infected part of the leaf. In the suggested model, the results, the classification accuracy, the segmented leaf image, and the fertilizer quantity for the diseases are presented with the index based on the type and severity of the diseases. In the suggested model, the results are presented based on the development of the end-to-end model, which is implemented as an interactive multilingual web application based on the Streamlit library. In the suggested model, the accuracy of the results exceeds 95% for the "PlantCNN" model, and the accuracy exceeds 97% for the "EfficientNet-B0" model, with the macro-averaged F1-score results exceeding 0.96 for all the test classes. The suggested model bridges the gap between real-world precision agriculture and high-quality deep learning research conducted in the lab.

Keywords: Leaf Segmentation; Precision Agriculture; Streamlit; Deep Learning; Convolutional Neural Network; Plant Village Dataset; Efficient Net; Transfer Learning; Plant Disease Diagnosis.