Study On Detection Method Of Wheat Ear In Field Based On Deep Neural Network

The automatic detection of wheat ears has certain research and application value in yield of estimation,seed screening,density estimation and gene trait expression.At present,most of the researches are in wheat ear counting.Aiming at the problems of low efficiency,subjective influence and lack of statistical research etc on the traditional function of wheat ears counting.This research has used the Shannongyoumai 2 wheat planted on Linjiatun Town,Zhangzhou City,Hebei Province(east longitude 115°99’,north latitude 39°43’)as the experimental object.Firstly,this paper has researched the automatic detection method of wheat ears based on YOLOv3 network,and has compared the detection results of data sets from different scale images.The datasets of different scale images were compared and analyzed.According to this result,this paper has researched out the automatic detection method based on Mask R-CNN network for the wheat ears,obtaining the projected area of wheat ears by mask.The results show that compared with the manual statistics,the recognition accuracy of wheat ears reached 87.12%,and the single sheet detection speed was up to 0.12s.Compared with the artificial statistics,the detection accuracy of wheat ears reached 97.00%,the single-sheet detection speed was up to 0.94s,and the relative error between the mask and the artificially-calibrated wheat ear pixel area was 11.61%.Two methods have certain accuracy,the YOLOv3 network detection is faster,the network Mask R-CNN is higher on detection accuracy,and the detection based on YOLOv3 and Mask R-CNN depth detection neural network for the wheat in the field is an relatively effective and accurate detection method.