Preprocessed Image Dataset of Indian Tea Leaves for Plant Disease Recognition: Helopeltis and Red Spider based

Present dataset contains a curated and preprocessed dataset of images of Indian tea leaves developed to facilitate research work in the field of plant disease recognition and automated classification using computer vision and machine learning techniques. The current dataset is dedicated to two important diseases of tea plants, i.e. Helopeltis infection and Red Spider and the healthy tea leaves, which constitute a three class classification problem under real field conditions. The dataset contains RGB images of the tea leaves from Indian regions of tea growing. Images were shot under natural light and background variations in order to maintain the complexity that occurs in the real world and between leaves in terms of leaf orientations, textures, and surface appearance. To improve the robustness and usability of deep learning models, extensive image preprocessing procedures were implemented and, as a result, several transformed versions of the original images were obtained. Three different classes are represented in the dataset: healthy tea leaves which have no known symptoms, tea leaves infested by Helopeltis with evident signs of puncture marks and localised necrotic areas, and tea leaves infested by Red Spider Mite with signs of bronzing, discoloration and surface degradation. Each class is stored as its own directory to enable the easy loading and labeling of models to be trained on the data.
A total of ten preprocessing techniques were implemented on each of the original images to produce a variety of feature representations. These include conversing the image to grayscale, converting the image to binary, Gaussian blurring with kernel sizes of 3x3, 5x5, 7x7, and 9x9, adding Gaussian noise with intensities of 2% and 4%, and rotating the images with an angle of +15deg and -15deg. The inclusion of multiple preprocessing variants means that the influence of noise, texture, intensity and geometrical transformation on disease recognition performance can be studied. All of the images are standard formatted and arranged to ensure consistency between experiments. The dataset can be used for starting tasks such as image classification, impact analysis of preprocessing, feature extraction comparison and benchmarking of deep learning architectures. By providing both original visual features and a variety of preprocessed features, the dataset is favourable for the establishment of robust and generalizable models for plant disease detection. This dataset is published for academic and research purposes and is intended to contribute on precision agriculture, smart farming system/sustainable tea crop management based on data-driven disease diagnosis contribution.