PestMango: A Multi-Source Dataset for Mango Pest Classification
Description
This repository provides the source provenance and documentation for the PestMango multi-source image dataset developed for mango pest classification. PestMango contains 18 pest classes divided into two subsets: a Global subset and a Gaza-Relevant subset. The accompanying files document the class information, image counts, and original data sources used in constructing the dataset. This repository supports reproducibility and transparency for research on automated pest classification using deep learning and hybrid CNN–Transformer architectures.
Files
Steps to reproduce
The dataset was constructed by combining pest images from multiple publicly available datasets. The classes were organized into Global and Gaza-Relevant subsets. Images were reviewed and assigned to their corresponding pest classes. Source information and class provenance are provided in the accompanying metadata files. For model experiments, the dataset was split into training, validation, and testing subsets using a stratified split. Image preprocessing included RGB conversion, resizing, cropping, and normalization. Data augmentation was applied only to the training subset.