Quality Detection And Sorting Of Soybean Seeds Based On Machine Vision

Soybeans are rich in various important nutrients and are the best protein sources for residents.In recent years,the demand for soybeans has increased,but domestic production has not significantly increased,resulting in a trend of expansion of the soybean deficit,which threatens China’s grain and edible oil security.There is an urgent need to improve soybean production and breed new soybean varieties.Seed selection is a critical step in breeding,since obtaining phenotypic information about soybean seeds and selecting superior varieties are essential for successful breeding.Soybean seed selection involves removing seeds from the plants after harvest,observing the phenotype of each seed,and recording the number and various phenotypic characteristics of the soybean seeds.Currently,soybean seed selection mainly relies on manual observation,which is inefficient and inaccurate and unable to quantify the phenotypic characteristics of soybean.To improve the efficiency and accuracy of seed selection,this thesis proposes an automated soybean seed quality inspection and sorting system based on machine vision.The main contributions of this thesis are as follows:1.An improved Mask R-CNN model suitable for detecting soybean seeds is proposed,in order to select defective soybean seeds accurately,based on a new feature extraction network and attention mechanism.The experiments show that the improved model increases the detection accuracy of defective soybean seeds by 19.92%compared to Mask R-CNN and improves the detection accuracy of all types of defective soybean seeds.2.A soybean feature parameter extraction algorithm based on watershed is used to obtain various phenotypic parameters of soybean seeds,including perimeter,area,and roundness.The algorithm can further sort soybeans based on quantified phenotypic parameters,thus further improving the accuracy of seed selection.The algorithm is embedded into the hardware platform,and the processor system and programmable logic are used to run the algorithm to improve the algorithm running speed.3.A soybean seed sorting device has been designed,which is equipped with the two soybean sorting algorithms mentioned above.By combining automation with mechanical equipment,the device is capable of using machine vision to sort soybean seeds.The system uses a conveyor belt to transport soybeans in batches,and a camera is used to capture images during the transportation process.The computer processes the captured images and uses the algorithm to sort out the target soybeans,the device then uses FPGA to control the corresponding pneumatic nozzle by opening the electromagnetic valve to remove the target soybeans,thereby achieving automatic sorting of soybean seeds.This thesis innovatively designs a soybean seed selection device based on artificial intelligence.The device integrates machine vision-based soybean seed phenotype feature acquisition algorithms into control system of seed selection,which significantly reduces the errors caused by manual seed selection,accelerates the breeding process,and improves the selection efficiency of modern seed industry.