Automatically detection of fruits on trees using deep learning-based object detection methods

Automatic detection of fruits on trees is a step towards increasing the productivity of production and estimating the yield of trees in agricultural management, including warehousing, number of workforce, category, etc. Automation is one of the harvest mechanization methods and machine vision has a pivotal role in this process. Therefore, it is essential to create an efficient model in natural daylight for fruit detection. In this study, 134 colorful images of orange trees has been employed. Since light conditions, fruit overlapping and clustering are the most important challenges in locating and detecting fruits in the garden, photography has been done in cloudy weather and at the night to solve the unfavorable sunlight problem. Then, three models of deep learning methods were employed to identify oranges. These models are: Faster RCNN ResNet101 ،SSD Inception V2 and Faster RCNN Inception V2. The Faster RCNN ResNet101 outperforms the others at 95.4% of mAP.

Keywords:

Object Detection , Deep Learning , Faster RCNN ResNet101