Research On Online Detection Of Rice Transplanter Operation Quality Based On 3D Reconstruction And YOLOv5s

Rice is one of the important food crops for human survival and plays an important role in Chinese agricultural production and national economy.Unqualified transplanting quality not only leads to the decline of rice yield and growth quality,but also leads to the reduction of economic benefits and the waste of seedlings.In the operation process of the existing transplanter,there are some phenomena,such as seedling lodging,floating seedling and leaking seedling.At present,there is no on-line testing system for transplanting quality that integrates seedling lodging,transplanting and transplanting.Therefore,this paper aims at detecting the upright degree of rice seedlings and identifying floating and leaking rice seedlings.Based on machine vision and image processing technology,it creates an automatic seedling measurement method based on double perspective stereo vision,constructs a rice seedling floating and leaking rice seedling detection model based on improved YOLOv5 s,and tries to produce an online inspection equipment for rice transplanter operation quality.It provides theoretical basis and equipment support for on-line identification and detection of rice seedling transplanting quality.The main research work and conclusions of this paper are as follows:(1)A kind of online inspection equipment for rice transplanter operation quality is designed.The equipment is mainly composed of image acquisition system,control system,seedling recognition system and visualization system.First of all,adjust the camera shooting Angle,and control the camera shooting seedling images according to the seedling transplanting rate;Secondly,the collected seedling images were introduced into the seedling recognition system,and the seedling uprightness,seedling drift and seedling leakage were counted.Finally,the visual interface is designed to display the transplanter transplanting quality.(2)An automatic measurement method of rice seedling uprightness based on two-Angle stereo vision was developed.Firstly,the seedling image was processed by image graying,and the seedling skeleton was extracted based on thinning algorithm and RANSAC algorithm.Secondly,the camera was calibrated,and linear fitting of seedling images from different angles was carried out based on double perspective stereoscopic vision to obtain seedling perpendicularity.Finally,under the significance level alpha = 0.01,one measurement work and the way to measure the significant difference of testing,the results show that: | t | < =0.22 t alpha / 2,this method is feasible,and with the artificial measurement there was no significant difference.(3)A rice seedling drift and leakage detection model based on the improved YOLOv5 s was constructed.Firstly,the images of floating seedlings and missing seedlings were amplified by contrast enhancement,brightness enhancement,random particle noise addition,etc.,to improve the generalization ability of the model.Secondly,the model trunk feature extraction network is optimized,Ghost module and EIOU loss function are introduced,CBAM attention mechanism module is added,the model is lightweight processing,in the suppression of background information interference,reduce the target positioning error and improve the accuracy of the model at the same time,reduce the complexity of the model.Finally,a control test was carried out,and the average accuracy of m AP@0.5 and m AP@0.5:0.95 were 93.03% and 80.92 respectively,which were 6.36% and 5.36% higher than the original YOLOv5 s model,respectively,meeting the actual work requirements of rice transplanting quality detection.(4)The online detection system of rice seedling drift and leakage was trialled.The algorithm and model proposed in this paper were integrated into the online testing equipment of transplanter quality,and the field experiment was carried out.The test results showed that the identification accuracy of drifted seedlings was greater than 90.75,and that of missed seedlings was greater than 90.73,and the evaluation indexes met the technical requirements of online testing of transplanter.