Research Of Lingwu Long Jujubes' Localization Based On Methods Of Scene Analysis

As an importantly representative and economic fruit,Lingwu Long Jujube belongs to Lingwu,a city in Ningxia,and it is also one of China protected geographical indication products.From mid-september to early October each year,Lingwu Long Jujube will usher in its picking period,which is concentrated.At present,ladder picking method and stick knocking method are the most commonly used method,but they are methods of manual picking,which will increase manpower and material resources,and they have low picking efficiency,which puts forward a strong demand for intelligent automatic picking.In the process,it is necessary to realize accurate identiflcation and positioning of Lingwu Long Jujube with the help of machine vision.The dissertation studies mainly Lingu Long Jujubes' Localization based on Methods of Scene analysis under natural conditions,which lays a theoretical basis for the harvest of Lingwu Long Jujube,and also provides technical support for the intelligent picking robot.The paper studies mainly from the following aspects:1.A single channel optimized noise reduction method based on HSI was proposed for the image preprocessing of Lingwu Long Jujube,and homomorphic filtering was used for image enhancement.The images collected in the experiment will be contaminated by noise.By converting RGB image into HSI image,According to that the three channels are uncorrelated mutually,and each component were optimized for noise reduction.Experiment shows that when the image contains specific noises,the improved algorithm is better than the traditional one;Under natural conditions,illumination has a great influence on the collected images.In order to keep the color invariance of image,homomorphic filtering method is proposed to reduce the influence of illumination.By converting images from the time domain to the frequency domain,and then using the stated transfer function to process the lighting component and the reflection component respectively,finally inverting to the time domain,the gray range of images was adjusted to reduce the influence of illumination.2.In order to achieve the separation of Lingwu Long Jujube and background,images of Lingwu Long Jujube were segmented.This paper presents an image segmentation method of Lingwu Long Jujube based on watershed algorithm.On the basis of traditional watershed algorithl,we used genetic algorithl to opti?ize the threshold,and the optimal threshold was obtained for image segmentation.Finally,Otsu and morphological operation were used for post-processing.Compared with the images obtained by manual segmentation,the segmentation accuracy of the improved algorithm is 89.99%;An image seglentation method of Lingwu Long Jujube based on scene analysis is proposed.The algorithm of K-means clustering was used to process images of Lingwu Long Jujube which had been converted into Lab model.Setting a specific scene with images of Lingwu Long Jujube,different K values were selected to segment the image.Finally,feature extraction and morphological operation were used to segment the image accurately.Compared with the images obtained by manual segmentation,the segmentation accuracy of the improved algorithm is 90.861%.3.Determining the number and centroid position of Lingwu Long Jujube in the image,and realizing the segmentation of the adhesion area of Lingwu Long Jujube.The paper presents an adhesion area segmentation method of Lingwu Long Jujube based on concave region.The segmented image was binarized and the region was marked.By closing each marked region and executing operation algebraically with the original image,the concave region can be obtained.The relationship between the number of concave regions and the number of Lingwu Long Jujubes was used to determine its numbers;In allusion to the condition that Lingwu Long Jujube stick to each other,using the corresponcKng concave region obtained,and line was the dividing line which was connected by the nearest two points under the Euclidean distance,then the line was used to realize the segmentation of the adhesion area of Lingwu Long Jujube;The centroid of the minimum outside rectangle of lingwu long jujube can be regarded as the centroid of Lingwu Long Jujube approximately,and then we can obtain the centroid position of Lingwu Long Jujube.4.The calibration of binocular camera and the determination of the centroid position of lingwu long jujube.The inner and outer parameters of the binocular camera were determined by the calibration method of zhang zhengyou,and the depth of Lingwu Long Jujube was obtained by parallax method,and the spatial position of lingwu long jujube was determined.The results of this dissertation can realize the localization of Lingwu Long Jujube and provide a theoretical basis for intelligent picking.