Application And Study Of Maize Leaf Epidermal Hair Identification Based On Improved YOLOv5

Maize is the largest,most productive,and fastest-growing food crop in China,and it is also an important element in determining the future food security of China,so research on maize varieties is a hot issue in academia.However,the distribution of epidermal hairs is dense and there are many small-sized epidermal hairs,which are time-consuming and error-prone to be counted manually.To address the above problems,this paper decided to use artificial intelligence to automatically identify and detect epidermal hairs on maize leaves and count the number of them to improve efficiency and accuracy.The existing common target detection algorithms have been experimentally proven to be ineffective in recognition,and there are false detections and omissions for some small target epidermal hairs recognition,so this paper proposes an improved YOLOv5 model to improve the detection of small target epidermal hairs and reduce false detections and omissions.The main research contents of this paper are as follows:(1)Data acquisition and pre-processing of maize leaf epidermal hair images.Because there is no publicly available data set for the epidermal hair images of maize leaves required for the study,the manual acquisition was performed.A total of 973 electron microscopic images of maize leaf epidermal hairs were collected at the top of each cob at 10 cm from the leaf tip and a total of 13,272 maize leaf epidermal hair study subjects were labeled with a resolution of 1024*943 pixels,due to the small data set and the data set was expanded to 4802 images using data enhancement,containing a total of 63,079 maize leaf epidermal hairs were studied.(2)To investigate the effectiveness of mainstream target detection models for the detection of epidermal hairs on maize leaves.Faster R-CNN,a representative model of two-stage target detection algorithm,and SSD,YOLOv3,YOLOv4,and YOLOv5,a representative model of one-stage target detection algorithm,were used to learn and train the epidermal hairs on maize leaves.The experimental results show that the traditional detection models have missed and false detection for small target epidermal hairs,and the YOLOv5 s model has better detection accuracy and speed in comparison,and the four model sizes of s,m,l,and x of YOLOv5 are compared and experimented,and the YOLOv5 s 6.0 model is finally chosen to be optimized and improved to solve the above problems.(3)A method of maize leaf epidermal hair identification based on the improved YOLOv5 model is proposed.The optimizer of the original training strategy is changed to Adam,and Label Smoothing is added.The C3 module in the backbone network of the YOLOv5 model is changed to the CAC3 module with the addition of the Coordinate Attention mechanism(CAC3),which is used to improve the localization ability of the model and focus more on small targets;asymmetric convolution is used to enhance the model for flipping and rotating objects,which can also provide better recognition of targets with different orientations,thus improving the detection accuracy of the model.The experimental results show that the proposed improved YOLOv5 algorithm has the best detection effect on corn leaf epidermal hairs,and the mean average precision(m AP)reaches 96.1%,which is higher than the average detection accuracy of Faster R-CNN,SSD,YOLOv3,YOLOv4,and YOLOv5 s 6.0 mainstream network model recognition methods for corn leaf epidermal hair recognition by 15.6%,29.6%,and 29.5%,respectively.improved by 15.6%,29.3%,11.9%,11%,and 1.3%,and the model processed each image in only 3.1ms more,and the average processing time of 10,000 images were only 31 s more.(4)A maize leaf epidermal hair recognition and detection system was designed and implemented.The interactive interface is designed based on Py Qt5 and Py Charm,and the improved corn leaf epidermal hair recognition algorithm proposed in this paper is adopted.The system test results show that the system can accurately identify the small and dense corn leaf epidermal hairs and accurately count the number of epidermal hairs,which can meet the practical application requirements.