Research On Maize Belt Recognition Based On Machine Vision And Path Planning

The application of machine vision in agriculture has changed the traditional way of agricultural machinery,liberated manual labor,and improved the efficiency of agricultural work.In view of the fact that the prevention and treatment of corn diseases in the middle and late stages of China is still mainly based on manual spraying agriculture,which is harmful to the health of the applicators,and the efficiency of agricultural sowing,fertilization,and pest elimination is low,this paper is based on image adaptive classification algorithm and volume-based The neural network method has carried out researches on the accurate identification of corn seedling belts,the accurate identification of corn rhizomes and path planning.The main work and conclusions are as follows:(1)Two image acquisition systems are designed according to the plant height of corn in different periods: the three-wheeled car structure mainly includes three wheels of the same size,the bracket system above the wheel structure,and the "T" structure with the camera installed.The tracked electric car mainly includes four parts: steering module,image acquisition module,power module and power module.The camera parameters and focal length are obtained through camera calibration.(2)Based on the designed two image acquisition systems,the collection of image samples in the critical period of corn growth is carried out.Seedling belts at the three-leaf stage and jointing stage are more obvious,and a sample bank of seedling belts and plants is established.The original images of plants in the three-leaf stage totaled 1,241 original images,with 106 min video of seedlings,the original images of plants in jointing stage totaled 1,453,and video of seedlings had 118min;the identification of seedlings in the trumpet stage and the tasseling stage was fuzzy,but the corn rhizomes were generally brown.Characteristics,establish a corn rhizome sample library.A total of about 6,000 images of the trumpet stage were collected,and the video duration of the rhizome was 180 minutes.A total of 5,200 images of the rhizome were collected during the tasseling period,and the video duration of the rhizome was 130 minutes.(3)Based on the image adaptive classification algorithm,the corn seedling belt is accurately identified and path planning is carried out,and G-R>(35)grand G-B>(35)gb color extraction operators are proposed for the green characteristics of the corn leaf belt at the three-leaf stage and the jointing stage,and based on K-means The clustering algorithm determines the method of determining the threshold of the color extraction operator indifferent environments.The path planning of the dynamic interest area is proposed,the vertical projection normal is used to determine the path of the static image,then the interest area of ? ? the static image is updated according to the position information of the recognition line,the interest area of the dynamic image is obtained,and finally the center of the corn seedling belt is fitted Line and navigation line,the deviation angle of the navigation line is 5 °,which greatly reduces the seedling pressing rate.(4)According to the demand for autonomous walking between corn rows of crawler electric trolleys,the accurate identification and path planning of corn rhizomes during the trumpet and tasseling stages were studied.Based on the pre-trained network model VGG-16,transfer learning was carried out,and a corn rhizome detection network was established.The model DOG pyramid algorithm was used to extract the target rhizomes in the image to form a sample training database,and the least square method was used to fit the path to the target rhizomes.The detection accuracy of corn rhizome recognition based on convolutional neural network reached 91.4%,much higher than the traditional detection accuracy.The research results of this paper have a certain reference value for the research of agricultural machinery based on machines.