Background and Objectives
The weight and size of crops are critical factors in determining their commercial value. However, manual measurement of these parameters is labor-intensive and time-consuming. The research aimed to develop a machine vision-based technology using a single RGB image to estimate the size and weight of Konjac corms in packed containers, addressing the limitations of manual measurements and the challenges posed by occlusion in packed conditions.
Methods
We proposed a novel method that employs DNNs to detect and measure Konjac accurately, even when obscured by others. A machine learning model was developed to identify occluded Konjacs and estimate their true dimensions by utilizing the features of surrounding Konjac. The method involved instance segmentation using CNNs, dimension estimation through ellipse fitting, occlusion identification using a binary classification model, and weight estimation using a regression model.
Results
Experimental results demonstrated a substantial reduction in measurement errors compared to traditional methods, highlighting the effectiveness of the proposed approach. The method achieved high accuracy in identifying occluded Konjacs and estimating their dimensions, leading to improved weight estimation accuracy. The system’s robustness and accuracy were validated in a field experiment, where it outperformed estimations made by a professional farmer.