Study On Health Identification And Transplanting Path Optimization Of Greenhouse Leafy Seedling

Modern agriculture needs intelligent transplanting equipment to replace manual transplanting of seedlings urgently.The machine vision assisted health assessment and path optimization of seedlings transplanting are key technologies to enhance the intelligence and efficiency of transplanting machines.However,machine vision recognizes of tilting pits,the recognition accuracy of seedlings and seedling transplanting path efficiency need to improve.Therefore,it is of great significance to develop an algorithm that can realize tilted tray correction and seedling health assessment and optimize the transplanting path to improve the automation of greenhouse transplanting.In this paper,an algorithm based on Canny and Hough transform is proposed to correct the image of tilted tray,and then evaluated the healthy state of seedlings in the image combined with the center of gravity,the area and the number of leaves.A new strategy combined seedling tray step control and greedy algorithm was proposed for transplanting optimization.Evaluated the recognition accuracy of the algorithm by tilted tray image,and built a transplanting process model to simulate the path optimization strategy.The main results obtained are as follows:Firstly,this paper studied the algorithm of tilt correction and location of seedling tray and extracted the key area of seedling tray.The contour line was detected by preprocessing,grayscale transformation,edge detection and Hough transform,the effect of different parameters were compared,and the deviation of alibration angle is less than 0.85 degrees.For corrected images,pixel projecting method were used to cut the tray area.Secondly,a health identify method of seedlings was proposed by evaluation of leaf area and the number of leaves,marked the healthy state and position of seedling.The main recognition object of this study were Salvia splendens seedlings which has been cultivated for two weeks.Seedlings were extracted and analyzed by preprocessing,binary transform and watershed algorithm.After the tray area divided,calculate the number and area of seedling leaf according to the gravity of each leaves,evaluated seedlings health status in each cells.The accuracy of identification of 200-hole seedlings with tilt angle within 1 degree was 99%.This solution provide health status and coordinates for the transplanting process.Finally,a step-by-step delivery strategy(SSDS)was developed for path planning.A total of 905 samples,including sparse trays and dense trays were used in simulation.The performances of SSDS,genetic algorithm(GA),greedy algorithm(GRA),and fixed sequence method(FSM)for planning the transplanting path were compared.The result shows,compared with FSM,SSDS shortening the transplanting path length by more than 16.5% in single tray transplanting and more than 9.8% in planning multiple-tray compound transplanting paths(MTCTP).For 128-hole trays with 26 vacant cells in the aim tray and 6 vacant cells in the transplanting tray,the average operation time of SSDS was 0.0016 s,which is shorter than those of GA(5.76s).Therefore,SSDS can satisfy the greenhouse factory requirements for continuous and real-time transplanting operations.