Abstract:
One of the most popular crops in Bangladesh is the bottle gourd. However, the quality
and productivity of the bottle gourd crop decrease due to a variety of diseases. Therefore,
a deep learning-based approach to identify disease is discussed in this study. We have
collected the dataset from agricultural field and applied various preprocessing techniques
like resizing, histogram equalization, augmentation etc. We have measured various
statistical values like PSNR, MSE, SSIM and RMSE in the dataset for the verification of
image quality after preprocessing the dataset. With the use of this research, farmers will
be able to spot bottle gourd leaf diseases early on, helping them to save money. Various
deep learning algorithms like VGG-16, MobileNetV2, CNN and DenseNet201 have been
used here. Using the dataset consisting of 1500 images of three classes (Healthy,
Anthracnose, Cercospora Leaf Spot), the models provided the accuracy of 83.33% for
VGG- 16, 93.33% for MobileNetV2 ,90.67% for CNN and 93.33% for DenseNet201.The
highest accuracy is provided by MobileNetV2 and DenseNet201 and it’s 93.33%.
