Deep Learning Recognition of Wheat Leaf Disease Using MobileNetV2 Model
Sara Bouskour, Mohamed Hicham Zaggaf, Lhoussain Bahatti
Abstract
Early stages are crucial in wheat production, as they contribute to minimizing crop losses in quality and quantity. Therefore, developing technologies becomes essential to accurately classify different categories of wheat leaf diseases, which plays a vital role in disease prevention. The constant advancements in Artificial Intelligence and computer vision have motivated researchers to use these powerful technologies for the automatic categorization of crop diseases caused by biotic and abiotic stresses in agriculture. Our study was based on two scenarios to conclude that MobileNetV2 represents the most appropriate choice amongst CNNs for the detection of wheat leaf diseases. Firstly, we used a 398-image dataset, with 70% for training, 15% for validation, and 15% for testing. Secondly, we used a larger dataset, totaling 1345 images distributed in 60% for training, 20% for validation, and 20% for testing. The conclusions of our analysis resolutely corroborated the undisputed efficiency of MobileNetV2. This model produced exceptional performance, with an F1-score of 0.99, an accuracy of 0.99, and a loss of 0.04. These remarkable results demonstrate MobileNetV2’s excellence in detecting wheat leaf diseases, resulting in high diagnostic acuity and low error propensity, regardless of the data volume used.