Detection and classification manipulator system for apple based on machine vision and optical technology

Abstract

Abstract. Classification is a significant step in the commercial processing of apple after harvest, however, at present, there are some problems in apple detection and classification, such as the simplification of detection index, relying on artificial classification and so on. With the development of robot technology, it has a good prospect to use manipulator to replace human labor to detect and classify apples. In order to realize automatic on-line detection and classification for apples, this study proposed an on-line detection and classification manipulator system for apples. Based on machine vision and optical technology, the system can detect the internal and external quality of apple at the same time. The system can divide apples into four levels referring to their internal and external quality. The manipulator consisted of an end effector and a six DOF mechanical arm, which can complete the grasping, transferring and placing of apple. The end actuator integrated the gripping mechanism and the near-infrared (NIR) detection module to realize the simultaneous grip and detection of the apple. The system used a CCD camera to get the image, and the image processing method was performed to get the apple position, diameter and other information after that. A spectrometer was used to collect the spectral data of 100 apples in 650-1100nm wavelength range. Spectrum collection and modeling in two different states of normal work state and step motor power off state were carried out, respectively. The SSC values of these 100 apple samples were measured immediately after that using the destructive method. And then the PLS regression model of SSC was established. Comparing the different modeling results, it showed that the griping process have no negative effect to the NIR acquisition process. Which illustrated the rationality of two process working at the same time. The results of SG smoothing and multiple scattering correction (MSC) preprocessing were that Rp = 0.9522 and RMSEp = 0.4492%. When the system works, the model can be invoked to predict the SSC value of apple. After testing the system function, the detection and classification function had been basically realized, which can replace the human labor to realize automatic detection and classification for apples. This system had stable and reliable performance, which can provide a new design method for fruit on-line nondestructive detection and classification equipment