On Tree Mango Fruit Detection and Counting System

Abstract

For yield estimation, it is crucial to achieve quick and precise identification of mango fruits in the natural situations and surroundings. Using imaging with computer vision to accurately detect and count fruits during plant growth is important. It is not just because it is a vital step toward automating procedures like harvesting but also for minimizing labour-intensive human assessments of phenotypic information which can be useful for the farmer. Fruit farmers or cultivators in agriculture would benefit greatly from being able to track and predict production prior to fruit harvest. In order to make the best use of the resources needed for each individual site, such as water use, fertiliser use, and other agricultural chemical compounds. Mango fruit is considered in this paper. A comparative study on Faster R-CNN, YOLOv3 algorithms, and YOLOv4 algorithms, which are widely used in the field of object recognition in the past on various fruits and objects, was conducted to find the best model. The YOLOv4 algorithm was chosen as it was the best technique for mango fruit recognition based on the findings of the above comparative study. A real-time mango fruit detection method utilizing YOLOv4 deep learning algorithm is put forward. The YOLOv4 (You Only Look Once) model was developed under the CSPDarknet53 framework. Also, the number of mangoes in the image or frame was counted and displayed in images as well as videos.