Abstract:
The article provides a unique deep learning strategy for automatic recognition of brinjal
plant diseases, using a large dataset that contains varied events of Phytophthora Blight,
Leaf Curl Disease, Withering Leaves, Late Blight, Healthy specimens, and Macrodiplosis
Dryobia. The objective is to increase early disease detection, allowing for immediate
assistance and improved crop management methods. Several machine learning algorithms,
including CNN01 Model, CNN02, VGG16, ResNet152V2, InceptionV3, DenseNet169,
and Transfer Learning Models, were examined to determine the best successful model for
this specific agricultural application.
To ensure models reliability and generalization, the dataset was thoroughly preprocessed,
including picture scaling, pixel leveling, and data enhancement techniques. The transfer
learning technique, which began with pre-training models on a large dataset like ImageNet,
aided in adapting these models to the unique features of brinjal illnesses.
Following thorough instruction and testing, the findings show that DenseNet169 is the bestperforming model, with an excellent accuracy of 99.77%. This accuracy outperforms
previous models, confirming DenseNet169's ability to properly categorize and detect
Phytophthora Blight, Leaf Curl Disease, Withering Leaves, Late Blight, Healthy states,
and Macrodiplosis Dryobia in brinjal plants. Because of its high accuracy, the suggested
deep learning technique offers great promise for real-world use in agriculture, providing
farmers with a strong tool for accurate disease detection and active crop protection. The
findings also highlight the necessity of selecting the correct deep learning structure for
plant disease detection applications that is suited to the details of the dataset.
