"INDEggPlantMSDD: A Multi-Scale Eggplant Surface Disease Dataset for Tiny and Large Lesion Detection Using Deep Transfer Learning"

"Early crop disease diagnosis is an important factor that enhances agricultural productivity and the quality of the vegetable crops in the contemporary farming mechanisms. Eggplant (brinjal) is among the most commonly grown vegetable crops in India and it is very prone to various surface level diseases due to the viral and environmental factors. These infections manifest themselves in the form of small spots, stains or spots on the surface of the fruits which in most cases diminishes the quality and market value of the crop. It is difficult to detect such symptoms at an early period since patterns of diseases can have different sizes, texture and appearance due to various conditions in the environment. In order to facilitate the creation of intelligent agricultural monitoring systems, the following dataset has been created: INDEggPlantMSDD: A Multi-Scale Eggplant Surface Disease Dataset based on Deep Transfer Learning to detect tiny and large lesions.The dataset consists of the images of eggplant fruit that were taken directly in the agricultural farms in Jaipur region of Rajasthan, India. The pictures were taken in the natural conditions of a farm in order to reflect the authentic changes in lighting, background surface, location of the fruit, and disease symptoms. The gathered raw images were then refined via various high-end image preprocessing algorithms to come up with a sound dataset that can be used to train deep learning models that can learn disease patterns at different spatial scales.The raw images were preprocessed and augmented using a total of 14 methods. Such preprocessing tasks are geometric transformations, like rotation at various angles, cropping tasks, Gaussian blur filters of various kernel sizes, injection of Gaussian noise, grayscale conversion, and other picture sharpening operations. Such transformations were meant to mimic different real world imaging conditions and raise the capability of machine learning models to identify the symptoms of diseases within the form of small spots and infections spread across the surface.The data is grouped into two major groups: Healthy and Rejected. The healthy category has the pictures of the eggplant fruits without any signs of diseases whereas the rejected category contains the pictures of the eggplant which have the surface disease or defect which renders it inappropriate to use in commercial purposes. The dataset has 2268 images in the healthy category and 2702 images in the rejected category after applying preprocessing operations.All the images were downsampled to a uniform resolution to allow them to be compatible with the current deep learning models. The multi-scale quality of the dataset is what allows it to be used to train more advanced models like convolutional neural networks, detection systems based on YOLO, EfficientNet and Vision Transformer based models.The INDEggPlantMSDD dataset should be used to conduct research on precise agriculture, automated vegetable quality inspection, and AI-based crop disease monitoring systems. Having a region specific dataset that is recorded in the true scenario of agriculture activities can assist in the creation of effective deep transfer learning models that are able to identify the existence of eggplant surface diseases in the real farming world."