The Detection Algorithm Of Navigation Path To Dwarf And Close Planting Jujube Orchard Based On Machine Vision Technology

The intelligent autonomous navigation technology of agricultural equipment can promote and improve the agricultural mechanization production efficiency.For jujube,Xinjiang province is one of the main producing areas in China,mainly using dwarf close planting mode by some relevant standards.In this way,offering a favorable application condition to machine vision navigation.In this paper,we put dwarf close planting orchard's plant protection period and harvest period mechanical operation as the research object,mainly discussed the distribution characteristics of the environmental target pixels in the two stages,the detection algorithm of the target characteristic discrete point group and by machine vision technology to fit visual navigation path as well.The main contents of this paper are as follows:Construction of dwarf close planting jujube orchard's visual navigation path detection system and image acquisition scheme.The software system consists of the target pixel distribution rule analysis module,the image preprocessing module,the path extraction module,the navigation path performance parameter extraction module.Hardware system image acquisition equipment selected Canon IXUS870 camera,image processing hardware platform used the processor AMD Athlon(tm)II X2 240 Processor,clock speed: 2.8GHz,memory: 2GB,system type: 32-bit WINDOWS 7 operating system computer.The detection algorithm of visual navigation candidate points to plant protection period and harvest period.For the plant protection period,the image was divided into two sub-categories to do,the first category includes that existing weeds among rows environment,fertilization and pruning environment,the second type is intertillage environment.For the first subclass,using the |R-B| method to transfer image into grayscale image,the second type adopted the traditional gray scale method,the two types used Ostu method to segment image,next used trapezoidal scanning algorithm,the area mark method,gray scale vertical projection method to denoise.For harvest image processing,classified into sunny,cloudy,backlight,smooth light,background multiple overlay five types to deal.Used B-component to gray scale,adopted the line-scanning auto-adaptive method to segment target and background.Through the inter-row scanning method,gray-scale vertical projection,morphological processing to denoise.Raised two trend lines to describe linear characteristics of jujube seedling,then used point-to-straight distance formula to extract the discrete candidate point group.The result showed that the extracted target point group is consistent with the distribution of jujube trees' stalks.The least squares method is used to fit the discrete candidate points to detect two edge lines,and then extracted the center line of two edge lines as the visual navigation path.The experimental result suggested that the algorithm of extraction path has high accuracy and robustness,and the detected characteristics points are consistent with the distribution of each period.In this paper,softer system processed image's resolution is 230×168.In the plant protection period,the average detection time of extraction path for each image is less than 14.0s,and the detection accuracy is higher than 78.3%,while video detection accuracy of more than 80%.In the harvest period,the average detection time of extraction path for each image is less than 9.2s,under single factor noise,the accuracy rate is higher than 83.4%,the multi-factors condition is 45%,and the average detection time of extraction path for each image is less than 9.4 s.while video detection accuracy is more than 81.3% under single factor,the average time taken for each frame is less than 1.7s,the accuracy of multiple operating conditions is 42.3%,and the average detection time is 1.0s.