Multi-Modal Rice Leaf Age Diagnosis Algorithm Based On Improved Residual Network

With the wide application of machine learning technology,modern agriculture gradually develops towards the direction of intelligent agriculture.The intelligent recognition of rice leaf age is an important part of intelligent agriculture.The intelligent recognition of rice leaf age firstly uses the image shooting equipment to shoot the images of the growth and development process of rice,and then uses the machine learning technology to extract the rice characteristic information from the rice image to predict the rice leaf age,and then makes effective regulation of rice planting according to the leaf age,so as to achieve the purpose of rice yield increase.This method saves time and effort and changes the traditional method of artificial leaf age diagnosis.However,at present,there are few intelligent leaf age detection algorithms.Some of these algorithms predict leaf age by single mode information such as rice image and accumulated temperature,while others predict leaf age by extracting rice leaf length combined with environmental information.The error is large,which hinders the popularization and application of intelligent system.Therefore,on the basis of residual network,rice image and leaf age diagnosis method,a new multi-modal rice leaf age diagnosis algorithm integrating unstructured data and structured data is proposed.The main work includes:(1)Aiming at the leaf age diagnosis of rice,a multi-modal data set consisting of rice image and rice growth environment data was constructed.Data sampling points were set in the farm,and the growth images of rice were collected periodically with image shooting equipment,and the unqualified images with noise and fuzzy picture quality were manually removed to construct unstructured data sets.At the same time,the average temperature,light intensity and other weather conditions in the growing process of rice were obtained from the weather station,and the accumulated temperature was calculated using the calculation formula of the average temperature and accumulated temperature.The sowing time and transplanting time were recorded manually,and the information of leaf age and leaf length of rice was measured regularly,and recorded on the field questionnaire to construct a structured data set.(2)A multi-modal rice leaf age diagnosis algorithm based on improved residual network was proposed.Structured data mapping network and residual network are taken as subnetworks to construct multi-modal data fusion network.Structured data mapping network deals with structured numerical features and residual network deals with unstructured rice image data.Firstly,the residual network was used to extract all the leaf morphology,color,texture and other information in the whole rice image.The rice image information output by the residual network and the high-dimensional structured numerical characteristic data output by the structured data mapping network were input into the multi-modal data fusion network,and the multi-modal data was used to predict the rice leaf age.The experimental results show that the error between the predicted value and the real value is about 0.5 leaf age.Compared with the single-mode method of leaf age prediction by only extracting rice image information through the residual network,the prediction accuracy is improved by about 0.3 leaf age,which has reached the standard of practical application.