Method And System Development Of Anthurium Grading Based On Machine Vision

With the rapid and steady development of our economy,agricultural economy gradually diversifies.Anthurium,as a kind of flower of great ornamental and economic value,is much loved by people.In order to ensure the maximum economic value of anthurium,anthurium needs to be identified by professional quality identification personnel before going out of the tent,and according to the quality of its graded pricing.Because of the large number of anthurium grading indexes and the direct contact and handling of anthurium in identification,the efficiency of artificial grading is low and it is easy to damage anthurium plants.In view of the above problems,the automatic classification system of anthurium is studied in this paper.The main research is as follows:(1)Investigate the research status of machine vision applied to crops,and introduce the related principles of image processing and target detection.Based on the anthurium grading index of Zhejiang Academy of Agricultural Sciences and the growth characteristics of anthurium,the grading index of anthurium and its parameter range were proposed.Based on Halcon calibration assistant,the internal and external parameters of industrial cameras were calibrated to obtain the mapping relationship between image pixel coordinate system and world coordinate system.The minimum enclosing rectangle algorithm based on the improved convex hull is proposed to measure the anthurium plant height.Compared with the minimum enclosing rectangle method,the algorithm is faster and the time is only 1/3 of the minimum enclosing rectangle.An elliptic fitting sum algorithm for anthurium crown size is proposed,which can better reflect the growth state of anthurium than the traditional crown size calculation method.The coverage of anthurium is calculated based on traditional image processing algorithm.Experimental results show that the proposed algorithm for measuring plant height,crown width and flower coverage meet the requirements.(2)Aiming at the characteristics of anthurium flame and pests,this paper proposes an improved YOLOv3 network model.Firstly,the FPN network of YOLOv3 is replaced by Bi FPN-3 network to realize the fusion of multi-scale and multi-layer features and improve the sensitivity field of the model.The channel attention module CAM is proposed,and the channel attention module is added after Bi FPN-3 output feature map,which can strengthen the effective features and reduce the noise.The decoupling detection head is proposed to replace the coupling detection head,so that the detection head can pay more attention to the task of classification and regression,and improve the accuracy and generalization ability of the model.The greedy NMS algorithm of YOLOv3 was replaced with Soft-NMS,which made the improved model have better detection effect against the obscured Buddha flame target.Experiments show that compared with YOLOv3,CSPYOLOv3 and other target detection models,the improved YOLOv3 model proposed in this paper has higher detection accuracy.(3)Design anthurium automatic classification system based on machine vision,including vision detection unit,motion control unit and classification execution unit.Design anthurium classification electrical control system,including the selection of PLC,electromagnetic relay,photoelectric sensor,motor,etc.,use STEP to write the ladder diagram of electrical control,and build the electrical control box.Combined with anthurium feature detection algorithm,using.NET Win Form platform and Halcon wrote the upper computer program to complete the software design of anthurium grading system.The experimental results show that the system has high grading efficiency and accuracy,and can be applied to the identification of large quantities of anthurium and meet the actual production needs.