Mango classification and segmentation dataset in Uganda

Mangoes are among the most important traditional fruits grown in Uganda, with several species cultivated across different regions of the country, including the Eastern region where the majority of our dataset was collected. Despite their economic and nutritional importance, limited work has been done on the classification and segmentation of local Ugandan mango varieties. This research aims to address that gap by developing reliable models for mango species classification and mango damage segmentation using a locally collected dataset.

The dataset is structured into three main components:

1. Mango Species Classification Dataset

This dataset contains 1,611 images of healthy mangoes, organized into nine different species, with each species stored in a separate folder. The images capture variations in shape, color, and texture across local mango varieties.

2. Segmentation Dataset

This dataset is designed for pixel-level damage detection and contains:

1,500 images of damaged mangoes

1,500 corresponding mango mask images

The segmentation images were augmented to increase dataset diversity and improve model generalization.

3. Damaged Dataset

This dataset is intended for damage/infection classification and consists of:

957 raw images of damaged/infected mangoes

6,419 augmented images generated to enhance model robustness

All mango images were captured under natural daylight conditions using smartphone cameras. The images were then preprocessed to remove background elements and resized to a uniform resolution of 256 × 256 pixels. All files are stored in JPG format.

This comprehensive dataset provides a strong foundation for developing machine learning and deep learning models for mango species classification and automated damage detection in local Ugandan mango varieties.