Lasso Regression Model of HSV Color Space Masking Vision Based Whole Coffee Bean Processing Agtron Measurement

The rapid increase in coffee consumption has led to a significant expansion in production scale and variety within the agricultural regions of the global coffee belt. Recent coffee harvested in varies specious and processed in various ways, and the functionality of processing equipment is advanced to optimize its quality. In particular, coffee roasting equipment has evolved to incorporate complex structures and advanced electronic, mechanical, and heat/air control technologies for processing harvested green coffee beans. One of the most critical quality factors in the roasting process is the degree of roast, which is measured by the bean color (Agtron) value. The conventional method for Agtron measurement involves grinding the beans and using vision sensors. However, the conventional method shows difficult and time-consuming. The method for measuring the color of whole beans without grinding is simpler, but it has limitations in accuracy due to variations in shading and background images. This study proposes a method to accurately measure the Agtron color of whole coffee beans without grinding. The proposed method employs a preprocessing technique that separates the measurement area from the background area through color histogram analysis. To enhance measurement accuracy, we suggest using a Lasso regression model, which mitigates overfitting problems. Experimental results demonstrate that the proposed method achieves an improved accuracy compared with traditional regression methods and neural network-based approaches.