Study On Rice Filling Stage Identification & Panicle Segmentation And Characteristics Analysis

Rice is one of the three main grains in China.At present,the judgment of the rice growth period is still artificial identification,which requires a lot of time and labor costs.The judgment standard of rice growth is not unified,and there is no quantitative index,so it is difficult for rice farmers to make scientific and reasonable field management strategies.Rice precipitates and sets seeds at the filling stage,and the field management strategy at the filling stage affects the yield and quality of rice.In order to accurately identify the rice filling stage and evaluate rice growth,a method of rice identification,segmentation,and characteristic analysis was proposed in this paper.Stage in the whole stages in rice,rice images using field camera acquisition,identify the filling stage of rice and then segment grouting period grain and grain characteristic parameters calculated,explore the relationship between the characteristic parameters and rice growing,the development of on-line monitoring and analysis software,the filling stage of rice for researchers and rice farmers,provide support for rice field management.The main work and results are as follows:1)Study on rice filling stage identification.At the beginning of the filling stage,some rice was still in the flowering stage,and it is difficult to identify rice at the flowering or filling stages because of their similar characteristics.According to the rice’s actual growth and development characteristics,rice images of the flowering and filling stage were spliced to construct a rice filling stage identification dataset.Compared with five classical convolutional neural networks,VGG-16 was finally selected as the rice filling stage recognition network model,and its recognition accuracy in the test set was up to 99.81%.2)Study on panicle segmentation of rice at the filling stage.It is difficult to segment the rice spike and leaves because they are interlaced and sticky.Therefore,based on the Deep Lab V3 Plus network model as the basic framework,the neural architecture search algorithm is used to automatically design the backbone network,modify the ASPP,and build Rice panicle segmentation network rice-Deeplab.Experimental results show that m Io U and Pix Accuracy of rice-Deeplab are 85.74% and 92.61% respectively,which are 6.5% and 2.97% higher than those of the original network model.3)Extraction and analysis of rice panicle characteristics at the filling stage.In order to explore the growth of rice,the characteristics of the rice panicle were designed,screened,and calculated based on the segmentation the of rice panicle image at the filling stage,and the relationship between the characteristics and growth of rice was analyzed.According to the experimental results,sparse or dense panicle,full panicle,green,golden,or gray color can be roughly distinguished by the proportion of panicle area,dispersion degree,average curvature,and color characteristics of the image.4)Design online monitoring and analysis software for the rice filling stage.Based on the Qt graphical interface library,the algorithm of growth period identification,rice spike segmentation,and parameter calculation is integrated,and the user-friendly human-computer interface is developed.The software can monitor rice growth online and provide quantitative parameters related to rice growth.