Combine Machine Vision Systems with Adaptive Neuro-Fuzzy Inference System to Classify the Maturity and to Locate the Stem Position of the Strawberry

• Main Authors: Yi-An Chen, 陳奕安
• Other Authors: Feng Ou-Yang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/21288936173186683516

碩士 === 國立宜蘭大學 === 生物機電工程學系碩士班 === 103 === Harvesting strawberries by robots will be the future trend to solve the problem caused by the shortage of the agricultural manpower. One of the critical factors for automatic harvesting is to find strawberries with sufficient maturity and their locations. The object of this research is to develop a machine vision system to identify the maturity of strawberries and locate their suitable stem positions for mechanical harvesting. This study used Saga-Honoka type strawberries for experiment. The images of strawberries were collected and the degrees of the maturities of strawberries were first classified by experienced farmers to build the database for further maturity classification. The degrees of the red colors on the fruit surface were categorized into three varieties of red colors, light red, middle, red and deep red. The area ratios of three red colors were counted using HSL color model. The traditional logic inference method was first used to estimate the maturities based on the area ratios of the red colors. The Adaptive Neuro-Fuzzy Inference System (ANFIS) method was then used to classify the maturities of the strawberry. First, ANFIS emulated the farmers to derive several ―IF-THEN‖ fuzzy rules. The learning method of the artificial neural network was then applied to adjust ANFIS parameters in order to improve their accuracy. The color pattern matching method was used to locate the stems for harvesting. The searching area for color pattern matching was constricted within the area above the fruits to shorten the time. The depth of the stem position was calculated by the binocular stereo vision formula. The result showed that the successful rate for maturity classification using traditional logic method was only 76%. The successful rate for maturity classification using ANFIS method can be increased to 93%. The result for comparing templates with different background ratios for color pattern matching showed that the template with about 30% background can locate the stems pattern with 94% successful rate. The binocular stereo vision can also successfully predict the depth of the distance between the stem and the camera. The developed system could be used to integrate with the robot arm to harvest the strawberries in the future.