Automated lag-phase detection in wine grapes using a mobile vision system

Abstract

Lag-phase in wine grapes refers to a critical stage in the growth cycle of berries when there is a significant slowdown in their growth rate roughly at half the harvest size. This stage is essential for predicting the crop yield at harvest as there is a strong relationship between the lag-phase berry size and the final berry size in grape clusters. This study is focused on developing an automated system to detect and predict the onset of the lag-phase in grape berries using a mobile machine vision technique. The system utilized various image processing techniques, including deep learning algorithms, to detect berries from grape clusters and estimate their size (diameter). The sizes of sample berries in randomly selected grape clusters were tracked and their growth patterns throughout the growing season were analyzed to identify the onset of the lag-phase. Berry size estimated using a Mask R-CNN-based segmentation and checkerboard calibration object achieved an RMSE of 0.473 mm and an R2 of 0.837 against the ground truth measurements. The mean absolute error (MAE) for detecting the lag-phase start date was estimated to be seven days. The results showed that the automated method was accurate and effective in detecting the onset of lag-phase. Overall, the study shows promise for using a deep learning-based technique to automate lag-phase detection and estimate crop yield, which can help growers make more informed decisions on crop management, inform harvest and post-harvest logistics, and improve the overall efficiency of farming operations.

Keywords

• Lag-phase
• Crop yield
• Mobile vision
• Image processing
• Deep learning
• Machine vision