Rice Phenotypic Feature Extraction And Yield Prediction Based On Deep Learning

Agriculture is an important cornerstone of China's economic development,and rice as the main food crop in China,with the growth of population,the consumption of rice is also gradually rising.In order to ensure food security in China,improving rice yield per unit area and cultivating excellent rice varieties is one of the important objectives of rice research.Rice panicle is an important part of rice,which determines the final yield of rice,so the research and analysis of Rice Panicle is very important.Accurate prediction of rice yield can accelerate the speed of breeding.The traditional rice yield prediction usually estimates the yield of rice by harvesting,threshing,drying and weighing rice samples in the field by field investigation method,which requires a lot of manpower and material resources,and is easy to cause human errors.The existing non-destructive rice yield prediction is mainly based on the spectral data of rice canopy and the growth environment factors such as temperature and rainfall.In this paper,a new method of rice yield prediction is proposed.Based on the rice image,the phenotypic characteristic parameters are extracted and the model is established to realize the rice yield prediction.The main research contents of this paper are as follows:(1)using deep learning segmentation network and image processing technology to segment rice in pot and field,get images of rice plant part and panicle,use deep learning detection and recognition network to calculate panicle number of potted rice.(2)On the basis of image segmentation,45 plant phenotypic features and 34 panicle phenotypic features of potted rice were extracted,32 plant phenotypic features and 33 panicle characteristics were extracted for field rice images.(3)Combined with the data of fresh weight and dry weight of panicle of potted rice extracted manually,and the yield data of field,the prediction model of fresh weight and dry weight of Rice Panicle was established by fitting the extracted phenotypic characteristic parameters.The determination coefficient R~2 of fresh weight prediction model reached 0.812,and that of dry weight prediction model reached 0.796.The determination coefficient R~2 of rice yield prediction model reached the highest of 0.683.