The Detection Of Citrus Surface Defect On The Tree Canopy At Nighttime Natural Environment Using Multi Light Sources

The traditional agricultural production is an important composition of China.In recent days,the labor costs have increased year after year.To adopt the machinery replacement,whose costs increase slower,can reduce the agricultural production costs effectively.With the rapid development of computer technology,information technology like artificial intelligence is maturing day by day.Combining agricultural machinery with information technology and artificial control,the construction of a new type agricultural machinery operating system will be the development direction of modern precision agriculture.Among many kinds of intelligent technologies,the machine vision is one of the fastest growing.By acquiring,analyzing and processing the image of the physical characteristics,the target can be recognized automatically.China is the largest producer of citrus in the world.Therefore,the visual location and defect recognition of citrus on the canopy can provide technic supports for citrus picking and yield estimation,which is effective to the accomplishment of precision agriculture and the cost control during production in our country.The research object of this paper is Calamondin citrus(Citrus microcarpa).The counting,defect recognition and classification of citrus on the canopy are conducted during nighttime in the natural environment.The research contents are as follows:1.Designed and produced the visual system for citrus images acquisition under two light sources in nighttime natural environment.Referring to the absorbance spectrum of citrus peel,the proper ultraviolet wavelength for the study was worked out by experiments.In order to obtain a stable lighting situation without much disturbance,the nighttime situation with less illumination was chosen.Both the white light and the UV light were integrated in the visual system and the switching control of light source was realized.Using this visual system,the images of citrus on the canopy were acquired in the same shooting position under different light source.2.The recognition and counting of citrus on tree was worked out in the nighttime environment.By analyzing the acquired white light image using different color models,the I component of YIQ color model was chosen for background removal,calculated by the Otsu algorithm.Then,the citrus image without background was used for edge detection and Hough circle detection.The amount of citrus and the fruit position was acquired.By comparing the recognition result with manual judgement,the counting precision is 98.6% and the recall rate is 91.7%.The result shows that the algorithm can count citrus fruit correctly.3.The identification and classification of citrus defects under ultraviolet light were carried out.Firstly,according to the circle detection result of the white light image,the background in the ultraviolet light image was removed,and the citrus fruit region was defined.The existence of defect was determined according to the peak number of the a component histogram from the Lab color model.An improved k-means clustering algorithm was then applied to the image containing the defect area.The citrus defect area with fluorescence excited by the ultraviolet light was segmented.According to the area difference between the scratched fruit and the shrinking fruit,which were manually classified and counted in advance,the classification standard was set.The defect was then divided into two types of scratching and shrinking.Two kinds of defect areas were used to made region grow calculation with the original citrus fruit area.The citrus fruit where the defect area was located was found out and counted in the ultraviolet light image.Finally,the identification and classification of the defect fruit were realized.Comparing the calculated results with manual judgment,the accuracy of defect fruit recognition is 99.8%,the recall rate is 92.2%,and the accuracy rate is 97.0%.For the classification of defective fruits,among 427 true positive samples,only one sample was classified incorrectly.The experimental results show that the recognition of fluorescence under ultraviolet light has a high accuracy rate for citrus defect recognition,which can provide technical support for citrus production.Combining multiple light sources and machine vision,the counting and quality recognition of citrus on tree were realized.The accuracy of the whole algorithm flow is 98.4%,and the average running time is 0.744 s.The results show that the proposed algorithm can detect the quality status of citrus on night trees effectively,and can provide technical support for citrus yield estimation and nighttime vision intelligence of picking robots.