| In order to meet the requirements of scientific management and yield estimation of appleorchards and to research the factors impacting its growth so as improve output and quality,this paper takes apples grown in BaiShui apple experiment station of our university asresearch objects, and studies the methods to acquire remote apple images and technologies todetect apple growth information.The main research contents and results are as follows:(1) This paper illustrates methods to acquire images of apples growing in trees andmeasure apples growth information based on the introduction of the remote images acquiringsystem. Meanwhile, the measuring system is designed according to the specific needs, and themarking and calibration methods referring to some references are ascertained.(2) To segment the apple from the image background, we conduct a study of segmentingapples from the image. Experimental results show that the segmentation method based oncolor characteristics cannot segment immature apple images from the background well innatural scene. So this paper proposed an image segmentation approach by combining textureinformation and graph theory based on minimum spanning tree. The image was firstly dividedinto many small image blocks, and the texture information of every image block wascalculated. Then the weighted function including the texture and location information weredefined by using image blocks as nodes and relations among image blocks as slides. To putthe distinctions among different nodes as weight, the undirected weighted graph wasconstructed and divided, before mapping the dividing results to the original image andobtaining the final image segmentation results. The experimental results show that thismethod has strong robustness, which has good results for segmenting immature apple images.The segmentation results are stable even with larger changes in light intensity, and theover-detailed division is improved.(3) Images segmented by the graph theory method were conducted secondary treatmentby the mathematical morphology method. The experiment indicates that the structure element with the size of5pixels was firstly carried on close operations and then open operations,which can fill the hole and reentrant and make the segmented fruit borders become moresmooth.(4) In light of the low speed of image segmentation methods based on the graph theory,we research and improve the fast template match algorithm based on image gray statistics.The collected images belong to the actual situation of image sequence, Firstly, the markedimage was used as the template on which gray statistic was conducted, and the cross lineartemplate was constructed by getting the max variance in horizontal and vertical direction.Then the detecting image was carried on gray statistic in order to reduce the search area. Atthe same time, using the pyramid algorithm for the reference and elimination mechanism, weutilize variable step method to optimize the relevance ranking algorithm for each matchedpoint. After fast locating the marked position in the image sequences and expanding thecurrent region, we segment the expanded image and conduct information calibration. Theexperimental results show this match method has a high speed, which can increasecomputation speed significantly.(5) This paper constructs a frame for detecting apple growth information including marksand achieves the image segmentation based on graph theory, morphologic process, the fastsegmentation of image sequences, the measurement of apple growth information onMATLAB platform. The experimental results show the relative error of longitudinal diameter,fruit shape index, and the size of fruit is within5%could fundamentally satisfy the accuracyrequirement in physical production. It is feasible to measure fruit growth information withoutdamages by analyzing images obtained in remote videos. |
PDF Full Text Request