Yield and maturity estimation of apples in orchards using a 3-step deep learning-based method

Abstract

Yield and maturity estimation of apples in orchards before harvest is essential for labor resource management. Current yield and maturity estimation are usually manually conducted, which is neither accurate nor efficient. In this paper, a 3-step deep learning-based approach for yield estimation and maturity classification is presented to address these issues. The proposed framework included an optimized detection network to count the visible fruits from both sides of a tree, a classification network to filter out mis-detected objects and perform maturity estimation, and a fruit load estimation algorithm to obtain the total fruit count of a tree. Images from three different apple orchards were collected to evaluate the performance of the proposed method. According to a series of comparative experiments, the proposed method outperformed several detection networks regarding the counting accuracy, indicating that an optimized architecture of the detection network combined with a fine classification network is necessary for enhanced precision of yield estimation. The presented workflow can be readily extended to other fruit crops for automation yield and maturity estimation featuring high efficiency and accuracy.