Automated recognition of optical image based potato leaf blight diseases using deep learning. (January 2022)

• Record Type: Journal Article
• Authors: Chakraborty, Kulendu Kashyap; Mukherjee, Rashmi; Chakroborty, Chandan; Bora, Kangkana
• Is Part Of: Physiological and molecular plant pathology. Volume 117(2022)
• Publication Date: 2022-01
• Subjects: Potato leaf -- Late blight -- Early blight -- Deep learning -- Parameter tweaking; Plant diseases -- Periodicals; Diseased plants -- Physiology -- Periodicals; Phytopathogenic microorganisms -- Host plants -- Periodicals; 632
• DOI: 10.1016/j.pmpp.2021.101781
• Languages: English
• ISSNs: 0885-5765

Abstract

Abstract: The Potato crop (Solanum tuberosum L. ) is one of the most important vegetable food crop grown globally. The yield of potato crop is greatly hampered both in quality and quantity by fungal blight diseases which pose a major threat to the global food security. Late blight caused by Phytophthora infestans and early blight caused by Alternaria solani are the most devastating foliage diseases for potato crops. In reality, the farmers presume such disorders by visualizing mainly the color change in the potato leaves that is usually risky due to subjectivity and huge time consumption. Under such situations, there is an urgent need to design computational models that would automatically detect these diseases rapidly and quantitatively even at its early phase. This paper explores recent deep learning models for automated recognition of late and early blight diseases based on the optical images of potato leaves. Initially, four deep learning models viz., VGG16, VGG19, MobileNet and ResNet50 have been trained with PlantVillage Dataset. It is observed that VGG16 provides the highest accuracy of 92.69% in comparison with other models. Now, to further enhance the performance of VGG16, fine-tuning of the model has been done based on the concept of parameter tweaking. The proposed methodology finally achieved 97.89% accuracy for classification between late and early blight syndromes as compared to healthy potato leaf. This study showed the detailed architecture of the fine-tuned VGG16 model with validation accuracy and losses. Our proposed methodology has also been compared with the existing techniques. Highlights: An AI based model is proposed for optical potato leaf image analysis for early and late blight disease identification. A benchmark dataset PlantVillage is used to perform this study. Analysis is being performed to compare four deep learning model namely VGG16, VGG19, ResNet50 and MobileNet. The best model VGG16 is selected and tried to enhance its performance by parameter tweaking.