| Summary: | 碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 93 === ABSTRACT
The purpose of this research is to quickly estimate fruit weight based on machine vision. The machine vision system includes light source, lighting, and image capturing, which transmits images to a personal computer via the IEEE 1394 interface. For the image we can distinguish objects from backgrounds using dominated colors for that fruit in the HSI model. The parameters, which involve the top area, side area, and volume size, are all estimated by machine vision techniques. Then a regression model between them and practical fruit weights is made. At last, the regression model is employed to predict fruit weight. This mechanism can be implemented in an automatic volume classification system to weigh fruits speedily.
In the experiment for weighing wax apples, we obtain the correlation coefficient of 0.937, RPD of 3.63, and the average error rate of 4.5% using principal component regression that the side area and volume serve as independent variables. In addition, we obtain the correlation coefficient of 0.936, RPD of 3.63, and the average error rate of 4.6% using simple linear regression that only the volume acts as the independent variable. In the experiment for weighing pears, we use simple linear regression to obtain the correlation coefficient of 0.936, RPD of 5.31, and the average error rate of 1.9% where the top area serve as an independent variable. Both of these experiments have good performance for classifying fruits.
By combining the machine vision techniques and regression analysis methods, our system estimating fruit weight only takes the execution time of 0.31 second. Such performance is better than conventional mechanical weighing machines.
Keywords: regression analysis, weight estimation, fruit, and machine vision
|
|---|
