Research On Recognition Method Of Pomelo Picking Point Based On Image Processing

Intelligentization and automation is one of the trend in research and development of agricultural equipment.The development of pomelo harvest machine could benefit the whole pomelo industry by increased effectiveness and reducing labor demand.Accurately identified the fruit and located the picking point at the stem were the basis for complying the function of the pomelo picking robot.In this work,the identification methods of pomelo fruit hanging on the trees in natural environment was proposed and tested.The recognition method was designed basing on image processing,which used fruit-hanging gesture and the centroid position to determine the region of interest and then determined the position of the picking point through the extraction of the fruit stem skeleton,used independent verification data to test the algorithm feasibility.The main research contents and conclusions were as follows:(1)The image acquisition system was built and the image data of the mature pomelo samples hanging on trees were obtained.213 pomelo images were collected,including 71 images under direct sunlight in sunny day,99 images under back lighting in sunny day and 43 images under cloudy day.The range of the distance between the camera and the fruit were from 0.2m to 0.6m.140 pomelo images under 3 kinds of lighting conditions were selected randomly to establish an algorithm model for pomelo picking points recognition.The left 73 pomelo images under 3 lighting conditions were randomly selected as independent verification sets to test the applicability of the recognition algorithm.(2)The method for identifying and segmenting pomelo was featured out.The segmentation performance of the chromatic aberration algorithm with chromatic aberration of 2,K-means clustering algorithm with clustering number K taking 2 and Cb color component were compared According to the results,chromatic aberration algorithm 2R-G-B color difference component composite image was selected for the pomelo fruit segmentation with combination with the OTSU adaptive threshold segmentation on the binary image of pomelo.The noise was eliminated through morphological operation and the method of removing larger area noise.The target of the divided pomelo could obtained by the optimized method.(3)An algorithm for pomelo picking point recognition was designed.Firstly the centroid and minimum circumscribed rectangle of pomelo was calculated.Secondly the fruit-hanging gesture was analyzed to determine the circular region of interest.Thirdly the V color component was used with 80 as the threshold,while,the noise was eliminated through morphological operations and the binary images of pomelo stems in circular regions of interest was extracted.Fourthly,the Zhang thinning algorithm was selected to extract skeleton from pomelo stem image.Finally,the method for calculating pomelo picking point was designed and the standard for obtaining the picking positioning was defined.(4)The picking point recognition algorithm was verified and the test results were analyzed.The influence of the radius of interest area on the picking point recognition were compared,excessively great or too low radius value could lead to misjudgment of picking points.The images of independent verification set were used to verifying the cutting point recognition algorithm.The average recognition accuracy rate was 87.67%.Among them,the average recognition accuracy rate of the 48 single pomelo images was 87.97%.The recognition accuracy rates under direct sunlight,back sunlight and cloudy day conditions were 88.89%,91.69% and 83.33%,respectively,with 0.33 s as average calculation time.The average recognition accuracy rate of 25 multiple pomelo images was 83.44%.The recognition accuracy rates under direct sunlight,back sunlight and cloudy day conditions were 85.71%,84.62% and 80.00%,respectively,with 0.36 s average calculation time.Analysed the positioning error,the misjudgment of the picking point was mainly came from the large average deviation of the pixels horizontal axis value.The research and experiment results show that the proposed recognition method of picking points based on image processing could figure out the target effectively.It had the potential to meet the requirements of the pomelo picking robot,which would provide a technical basis for pomelo picking robot design and control.