| 【Objective】 The purpose was to provide a fast and accurate method for the identification and counting
of mature citrus in natural environment, to solve the shortage of high cost, long time and low precision caused by manual
sampling method, and to lay a foundation for automatic picking of citrus in the future. 【Method】 The RGB camera was
used to collect the image of the citrus fruit tree, and the image was converted to Lab color space. The“a”component was
used for the citrus distinguishing from the background color, and then the MATLAB software was used to count the citrus
based on Hough circle transformation method to achieve an estimate of the citrus yield. 【Result】 The image processing
method is simpler and faster than the traditional method of fruit and background separation. The recognition accuracy
rate is 94.01%. Yield estimation accuracy is 96.58%, and the average recognition time is 1.03 seconds. The algorithm
was tested on 20 images(10 trees), and the number of fruits counted by this algorithm was compared with that counted by human observation. The coefficient of determination (R2) is 0.83. 【Conclusion】 The method can realize rapid and automatic
identification and counting of fruits and has good robustness to fruit overlap and fruit occlusion. This research promotes the
application of machine learning in modern agriculture, has a high theoretical and practical significance, and facilitates the
further development of orchard smart agriculture. |
