Study On Rapid Detection Of Navel Orange Surface Defect Based On Machine Vision And Embedded System Application

Defect detection is an important part of fruit grading.Due to the diversity and complexity of fruit surface defects,the rapid detection of fruit surface defects is the research focus in the field of academia and industry.In recent years,computer vision technology has been applied to the external quality inspection of agricultural products.In this paper,the detection methods of various types of defects on navel orange surface(e.g.,thrips scarring,canker spot,dehiscent fruit,copper burn,phytotoxicity,insect injury,wind scarring,scale infestation)using machine vision technology were proposed.An automatic defect detection experiment system for navel oranges was also developed.The proposed detection methods offered help for developing an on-line navel orange defect detection system.Main contents and results in the paper are listed as follows.(1)The defect detection platforms using major PC-based machine vision system and the low-cost embedded vision system were originally developed.(2)Automatic detection of defective oranges is not easy because of the uneven lightness distribution on the surface of navel oranges.A novel fast multi-threshold gray-level edge segmentation algorithm for defect detection was proposed.This method was an original contribution that allows successful segmentation of various types of surface defects.In order to evaluate performance of this algorithm,5008 regions of interest from 8 types of defective orange samples(e.g.,thrips scarring,canker spot,dehiscent fruit,copper bun,phytotoxicity,wind scarring,insect injury,scale infestation)were marked to test.The method achieved the performance rate of 92%successful segmentation results on individual defects.(3)A novel fast local gray-level segmentation algorithm based on integral image concept for defect detection was proposed.This method was an original contribution that allowed successful detection of various types of surface defects when navel orange images presented faint defect characters or inhomogeneous surface.The method was also tested by various types of static defective orange images(e.g.,thrips scarring,canker spot,dehiscent fruit,copper burn,phytotoxicity,wind scarring,insect injury,scale infestation).The classification success ratio was 95.2%with a processing time 38.5 ms per orange image.(4)A novel fast adaptive lightness correction method using single threshold segmentation algorithm defect detection was proposed.The true defect region on fruit surface was still in relatively lower intensity,and the sound peel was stretched up to high intensity,so that it was more easier to segment the defect regions accurately using a simple global threshold.The algorithm was tested by static defective orange images in the different lighting conditions from the published research papers.This study also compared other lightness correction algorithm implementations for defect detection.The proposed adaptive lightness correction method was more than 10 times faster than the existing methods.(5)An online detection system for navel orange surface defects using a low-cost small embedded machine vision based on ARM processor with an industry gigabit ethernet camera was developed.The work included the embedded system software and hardware development,online imaging environment development,embedded image acquisition,Linux embedded system driver software design for the gigabit network industrial camera,embedded algorithm for image processing,embedded software architecture and design for defect detection,etc.The online system performance was tested with the most common surface defect of navel oranges.The classification success ratio of navel oranges was 95.8%with a processing performance of 7 oranges in one second using a single online detection channel in the small embedded machine vision system.