Design And Experiment Of Automatic Film Breaker Based On Deep Learning Potato Seedling Detection

Plastic film mulching cultivation is a widely used technology in potato planting industry.Mulching has the effects of warming,water retention,fertilizer retention and weed growth inhibition,which can effectively improve the yield and quality of potato.In order to ensure the normal growth of potatoes,it is necessary to break the film covering layer in time so that the seedlings can get sufficient sunlight and oxygen,and at the same time,it can avoid the problem of too high or too low soil temperature caused by too long film covering time.At present,the membrane breaking of potato seedlings mainly depends on manual work.In view of the high labor intensity and low efficiency of manual membrane breaking,this paper proposes a method of using image detection to identify potato seedlings and driving the membrane breaking knife through the mechanical arm to break the membrane.An automatic membrane breaking machine based on deep learning potato seedling detection is developed.The main research is as follows :(1)Overall design of the machine.According to the working environment of the film breaker,the mechanical structure and control system of the whole machine are determined.The mechanical structure includes frame,film breaking motion platform,self-propelled drive system,seedling identification mechanism and film breaking device.The control system system includes seedling identification system,motion platform membrane breaking system and self-propelled drive control system.(2)A potato seedling detection method based on YOLOX network was proposed.Firstly,the potato seedling image sample data set was constructed.The potato seedling image was collected according to the test requirements.After data enhancement and image annotation of the collected seedling samples,the YOLOX network was used to train and detect the seedling samples.By continuously optimizing the network structure and adjusting the hyperparameters,the optimal detection model was achieved.The experimental results show that the average accuracy of the network for potato seedling recognition is 97.26 %,the return rate is 96.25 %,and the harmonic mean F1 is 0.91.The model can quickly and accurately identify potato seedlings.(3)The camera is calibrated to eliminate the error caused by camera distortion.The hand-eye calibration of the camera and the moving membrane breaking platform is performed to obtain the conversion relationship between the pixel coordinates and the platform coordinates.The seedling recognition system outputs the pixel coordinates of the seedlings in the camera coordinate system while identifying the potato seedlings,and converts the pixel coordinates into the position coordinates based on the membrane breaking platform.The position coordinates are sent to the lower computer through serial communication,and the membrane breaking operation is carried out according to the set membrane breaking strategy.(4)The field experiment of potato seedling membrane breaking was completed.Firstly,the working conditions of the field test were determined,and the assembly and debugging of the test prototype were carried out.Secondly,the evaluation index of the membrane breaking performance of the automatic membrane breaking machine was determined.Thirdly,the membrane breaking operation of potato seedlings was carried out through the workflow determined in the previous chapters.Finally,the membrane breaking data of the automatic membrane breaking machine were analyzed.The results showed that the qualified rate of membrane breaking was 91.6%,the seedling damage rate was 3.4%,the accuracy of seedling identification was 95%,and the membrane breaking efficiency was 11.32 pieces/min.The basic completion of the seedling membrane breaking work provides a strong guarantee for the subsequent growth of potato seedlings.