BottleNet18: Deep Learning-Based Bottle Gourd Leaf Disease Classification

Abstract

Plant disease classification is often accomplished by visual assessment or during research facility assessment which creates setbacks bringing about yield in loss when diagnosis is completed. Plant disease detection through an automated approach is advantageous because it minimizes the amount of monitoring required in large crop farms and identifies disease signs at an early stage, i.e., when they develop on plant leaves. Our suggested method adds to the automatic recognition of plant diseases through a series of processes that include pre-processing, analysis, and classification. In this study, an unsharp masking filter utilizes to process the blurred and the unsharpened part of the real images presents as a mask for producing a sharpened resulting image. As an image enhancement, a green fire blue filter is used to enrich the quality of images by increasing the contrast, removal the colors, and thresholding the images. For the verification of image quality, several statistics formulas such as PSNR, MSE, SSIM and SNR are calculated in the dataset. And finally, a proposed bottlenet18 deep learning architecture has been applied to classify three different Bottle gourd diseases as Anthracnose, Cercospora leaf spot, and Powdery mildew. In this work, we have measured the performance based on the performance matrices with variations of different optimizers and learning rates. The highest accuracy achieved by using the proposed BottleNet18 architecture is 93.9987% with Adam optimizer and 0.001 learning rate.