Detection of Strawberry Flowers in Outdoor Field by Deep Neural Network

Abstract

Listen

Translate article

This paper proposed an accurate, fast and reliable strawberry flower detection system for the automated strawberry flower yield estimation and harvesting. A state-of-the-art deep-level object detection framework of region-based convolutional neural network (R-CNN) was developed for improving the accuracy of detecting strawberry flowers in outdoor field. The networks were trained on 400 strawberry flower images and tested on 100 strawberry flower images. To capture features on multiple scales, three different region-based object detection methods including R-CNN, Fast R-CNN and Faster R-CNN were presented to represent the strawberry flower instances. The detection rate for R-CNN, Fast R-CNN and Faster R-CNN models were 63.4%, 76.7% and 86.1 %, respectively. Experimental results showed that the Faster R-CNN method archives better performance than R-CNN and Fast R-CNN and is less time consuming. We demonstrated the performance of the Faster RCNN framework even if strawberry flower are occluded by foliage, under shadow, or if there is some degree of overlap among strawberry flowers. Moreover, automatic yield estimation provides a viable solution for the current manual counting for yield estimation of fruits or flowers by workers which is very time consuming and expensive and also not practical for big fields.