MangoFruitDDS: A Standard Mango Fruit Diseases Dataset Made in Africa

Mango is a lucrative fruit produced in tropical and sub-tropical areas. It is the third most traded tropical fruit after pineapple and avocado in the international market. In Senegal, the average production of mango fruits between the 2015–2016 and 2021–2022 seasons is 126,551 tons. Mango fruit is also leading the fruit exportation of the country. For example, in the 2017–2018 season, the quantity of mangoes exported was estimated at 17.5% of the country’s fruit production, ahead of melon (13.4%) and watermelon (11.6%). There are, therefore several pests and diseases that reduce both the quantity and quality of mango production in the country. Several solutions based on Convolutional Neural Networks (CNNs) are proposed by researchers during the last years to automatically diagnose these pests and diseases. But the main limitation of these solutions is the lack of data since CNNs are data-intensive. Due to climatic variations from one geographical area to another, these solutions can only be adapted to certain areas. We propose in this work a mango fruit diseases dataset of 862 images collected from an orchard located in Senegal. Two combinations of data augmentation techniques, namely “Flip_Contrast_AffineTransformation” and “Flip_Zoom_AffineTransformation” are used to generate respectively two datasets: Dataset1 and Dataset2 of 37,432 images each one. Eight CNNs, including seven well-known ones and a proposed light weight Convolutional Neural Network (LCNN), are applied to both datasets to detect and identify the treated diseases. Results show that on Dataset1, DenseNet121 and ResNet50 give the best accuracy and F1_score both equal to 98.20%, on Dataset2, InceptionV3 and MobileNetV2 achieve both the best accuracy and F1_score of 98.20%. The proposed LCNN also achieved excellent results (accuracy: 95.25% and F1_score: 95.20%) on dataset1. Due to its light weight, it is therefore deployed in an offline Android mobile application to help users detect mango diseases from captured images.