Recognition And Localization Of Papillae In Millet Cells Based On Deep Learning

Millet is one of the important economic crops in our country and is planted in large quantities in Shanxi Province.A large number of studies have shown that the formation of cell papillae is related to plant disease resistance,so the collection of millet cell papillae information is of great significance for the study of millet disease resistance.The traditional method of collecting papillary information is manual counting,which is inefficient and has errors.In this paper,deep learning is used to identify and locate millet cell papillae and collect information on millet cell papillae,so as to improve efficiency and reduce errors,which is of great significance for millet disease resistance research and millet breeding.This paper uses Faster R-CNN algorithm based on Inception Resnet v2 and Inception v2 feature extraction network and SSD algorithm based on Inception v2 and MobileNet v1 feature extraction network to conduct experiments.The three feature extraction networks of Inception v2,Inception Resnet v2 and MobileNet v1 used in the experiment were elaborated and analyzed,and the framework of Faster R-CNN and SSD target detection algorithm was studied.In the experiment,the SSD Inception v2 model and Faster R-CNN Inception Resnet v2 model lost too much during the training process.When IOU=0.5,the average accuracy is less than 10%,and the millet cell papillae can hardly be detected,while Faster R-CNN Inception v2 The average accuracy of the SSD MobileNet v1 model is above 55%,which can effectively detect the target.Comparing the results of the Faster R-CNN Inception v2 and SSD MobileNet v1 model experiments with manual counting,it is found that the Faster R-CNN Inception v2 network model is better than SSD MobileNet v1 in the detection of mastoids in the identification and localization of millet cell papillae The speed of the network model is faster and the average accuracy is high.Especially in the detection of small mastoids,Faster R-CNN Inception v2 has a much better detection effect than SSD Mobile Net v1.Compared with manual counting,the Faster R-CNN Inception v2 network model can more accurately identify and detect the papillae in the picture,and the detection of millet has the best effect and high reliability.Although Faster R-CNN Inception v2 network model has better detection effect and speed,there is still room for further optimization of Faster R-CNN Inception v2 network model.The accuracy of the network model can be improved by expanding the data set and adjusting network configuration parameters.This paper designs and develops an automatic mastoid detection system based on the Faster R-CNN Inception v2 network model.It realizes multi-scale identification and localization of cell mastoids in pictures,and counts the number of cell mastoids of different sizes in a single image.