Research On Autonomous Path Detection Method For Orchard Carrier Platform Based On Convolutional Neural Network

Orchard planting industry occupies a large part in China’s agricultural economy and is an important industrial pillar for local poverty alleviation in China.In recent years,with the development of information technology,agriculture is moving in the direction of information and intelligence.And mechanical intelligent level of the orchard of our country is low,in most of the fruit crop production is still done by traditional experience homework between garden,the orchard in the production of bagging,fertilization,pruning,spraying,picking the intensity of labor is larger,a variety of orchard machinery in different seasonal perform different tasks,equipment utilization rate is very low,the adapter is difficult.It is imperative to promote intelligent machinery equipment suitable for such operations.Carrying machinery of different operation links through public transport platforms will greatly liberate labor,improve production efficiency and reduce economic costs.In this paper,the autonomous detection method of the path of the orchard carrier platform based on convolutional neural network was proposed.The trunk and obstacle detection were carried out through the target recognition model,and the autonomous detection of the path of the carrier platform was completed according to the coordinate information obtained from the position of the trunks on both sides.The main contents and conclusions of this study are as follows:（1）Taking the orchard as the research object,according to the characteristics of the semi-structured environment of the orchard,the design scheme of the path autonomous detection method of the orchard carrying platform was determined.Firstly,image samples of orchards in different seasons and under different light conditions were collected to better simulate the operating environment of orchards in different seasons.Standard data sets were constructed and target recognition models were obtained by using improved YOLOv5 s network training.The memory size of the model is 13.99 MB,and the recognition speed is 33 ms.The accuracy and stability of the model are tested through the test set,and the average accuracy of the model is 95.2%.The experimental results show that the model is suitable for semi-structured orchard environment.（2）The trained target recognition model is combined with the carrier platform to realize path detection.When the orchard carrying platform is moving in the orchard,the improved YOLOv5 s target recognition model is used for real-time detection,and the real-time coordinates of the target frame are printed out to obtain the position information of the recognition frame.The next target point of the path is obtained by fitting the centroid coordinate points of the recognition frame with the least square method.The test results of the carrier platform show that the carrier platform,moving at a speed of 0.7m/s,can detect the path at the same time as the target identification,which can provide a path detection method for the orchard to be applied to the actual environment.（3）Based on the upper computer to draw the path map,pixel path information.The pixel path information map planned by A* algorithm is displayed on the interface of upper computer,and the calculated results show that the shortest path is found while avoiding obstacles.The actual path coordinates are obtained by converting the pixel path coordinates to the world path coordinates,which provides support for the autonomous detection of the path of the orchard carrier platform.