Citrus Leaf Disease Classification Using Different Deep Learning Models: A Survey of Southern part of Bangladesh

Citrus plant cultivation is becoming increasingly popular in South Asia, more specifically in India and Bangladesh region due to its rapidly growing demand as well as suitable weather conditions for farming. It can be affected by several diseases and requires keen attention to detect and cure the diseases in time; otherwise, significant monetary loss is incurred. With the advancement of computer vision and deep learning techniques, identifying various diseases is becoming simpler. However, this process requires a proper dataset of infected leaves and a suitable detector to recognize the diseases, and the existing renowned citrus disease datasets are based on a particular zone that may not be like the actual data that apply to the Barishal region. Therefore, a new dataset is proposed by doing two-step surveying in the southern part of Bangladesh that comprises images of infected citrus leaves with multiple classes of diseases i.e. 5 types of diseases, including precise annotations. Our subsequent investigation with these datasets consists of selecting basic CNN models for classification as well as applying the dataset to existing deep learning models such as VGG16, Xception, and InceptionV3 for performance and computational analysis. The high performance of those models proves the potentiality of the dataset. The dataset's diversity, clear images, transparent backgrounds, and real-life weather conditions led to the models achieving remarkable accuracy.