Gradient Boosting Optimization using Tree-structured Parzen Estimator for Estimating Tomato Weight in Conveyor Systems

Abstract

Automated fruit sorting is a critical component of smart agriculture, facilitating efficient and precise classification of fruits based on various parameters. This research proposes the use of Gradient Boosting optimized with Tree-structured Parzen Estimator (TPE) to improve the accuracy of tomato weight estimation. The data collection process involved the use of a conveyor system, in which tomato videos were recorded using a vertically mounted camera at conveyor speeds of 40 RPM, 60 RPM, and 80 RPM. This approach was adopted to assess the impact of speed on the accuracy of prediction. The Gradient Boosting model was employed to predict the weight of tomatoes, and its performance was compared with the optimized version using TPE. The TPE method involves a search for the optimal hyperparameter configuration, leading to a reduction in loss value compared to a model lacking optimization. The experimental findings demonstrate that at a conveyor speed of 40 RPM, the TPE-optimized model achieved a maximum $\mathbf{R}^{\mathbf{2}}$ of 0.95, while at 60 RPM and 80 RPM, the $\mathrm{R}^{2}$ values decreased to 0.92 and 0.89, respectively. This decline in accuracy at higher speeds can be attributed to the increased blurring of the image due to the accelerated movement of the objects. Furthermore, the Gradient Boosting model that is not optimized with TPE produces lower accuracy, with an average error approximately 5 to $10 \%$ higher than the optimized model.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint