CNN-based Early Blight and Late Blight Disease Detection on Potato Leaves

Potatoes are grown commercially in practically every country in the world. Unfortunately, the crop has been affected by a number of different diseases. In order for the gardener to take quick action, they need to have an understanding of the nature of the contamination. They had the notion that if they looked closely at the leaves, they would be able to learn more about the diseases that were plaguing their communities. Many different Convolutional Neural Network (CNN) models and Machine Learning (ML) methodologies have been created in order to provide assistance to farmers in the diagnosis of diseases affecting tomato crops. Deep Learning and Neural Networks are used in the construction of CNN models. This gives CNN models an advantage over other Machine Learning approaches, such as k-NN and Decision Trees. Because it must handle such a wide array of inputs, the notoriously challenging Pre-skilled CNN is notoriously tough to programme. However, it is capable of producing incredible works of art. An outline of a model for a convolutional neural network that is simpler to understand is provided here. It consists of a total of eight hidden levels. The suggested lightweight model beats both state-of-the-art machine learning approaches and pre-trained models in terms of accuracy when applied to the Plant Village dataset, which is available to the general public. The Plant Village dataset has 39 classes, and these classes collectively represent a large number of different plant species. There are ten different diseases that may infect tomato plants, all of which have the potential to inflict damage. While k-NN has the best accuracy (94.9%) among the classic machine learning methods, VGG16 performs exceptionally well among the trained models. After the picture improvement was finished, the images were pre-processed so that the effectiveness of the suggested CNN may be increased. To be more specific, we accomplished this by considering the width of the picture as a random variable and, as a result, altering the brightness of the image correspondingly. On data sets that have nothing to do with Plant Village, the suggested model achieves an outstanding accuracy of 98%.