The Research On Estimation Methods Of Chlorophyll Content In Wheat Based On UAV Aerial Images

Winter wheat is three major food crops in China,and its production status directly affects my country’s food security.Wheat growth monitoring technology based on UAV remote sensing images plays an important role in modern wheat yield and management.Chlorophyll is an important pigment for photosynthesis in plants,and its content is closely related to the growth status and nutritional status of crops.Therefore,the chlorophyll content(SPAD value)has become an important index for evaluating the growth status of crops.The traditional process of monitoring wheat SPAD value is time-consuming,laborious,complicated,and destructive.Therefore,it is an important measure to carry out efficient,non-destructive and remote sensing monitoring of SPAD value of large-area winter wheat that is suitable for realizing winter wheat growth monitoring and yield evaluation.When monitoring wheat growth,UAVs are equipped with vision cameras to obtain wheat field images,which are flexible in operation,cheap in price and more efficient.The realization of winter wheat SPAD value by UAV remote sensing quantitative estimation and remote sensing inversion is the research goal,and related research work is carried out.The research results are as follows:(1)Semantic segmentation of UAV wheat image based on deep learning model is studied.SegNet segmentation model was applied to wheat canopy image segmentation based on UAV,and compared with traditional segmentation methods.First of all,LabelMe software is needed to make label graphs and enhance the acquired data.Secondly,various parameters of SegNet segmentation model are optimized.Finally,the results of the mask segmentation were compared with the OTSU,the maximum entropy threshold method and the K-means method.The results showed that the segmentation effect of SegNet algorithm is the best.The pixel accuracy of the SegNet model reaches 93.98%,and the Dice coefficient up to 92.38%.Experiments show that based on the drone technology,the SegNet model is more suitable for the segmentation of wheat canopy images obtained by drones under this parameter(the activation function is ReLU,the learning rate is 0.0001,and the number of iterations is 500),which is similar to the traditional segmentation method.In contrast,it has strong robustness.(2)The feature extraction of five different color spaces was studied.Based on SegNet algorithm segmentation of wheat canopy image,the R,G,B,H,S,I,Y,U,V,Y,Cb,Cr,L*,a*and b*components of RGB,HIS,YUV,YCbCr and L*a*b*five color spaces were extracted by image processing technology.(3)A prediction model of wheat canopy chlorophyll content based on traditional machine learning and deep learning was built.The wheat canopy chlorophyll prediction models of support vector regression(SVR),gray wolf optimization support vector regression(GWOSVR)and convolutional neural network regression(CNNR)were constructed separately based on the extracted single color space component and multiple color space components,and CNNR models with different structures were designed.The experimental results show that in the component prediction effect based on single color space,the components in L*A*B*space are the best and the components in RGB space are the weakest.Based on the prediction effect comparison of five color space components,the constructed CNN 1D-4 model based on the four-layer convolutional layer has the highest prediction accuracy,and the determination coefficient of its test set is as high as 0.875,which is 37.1%higher than support vector regression and 17.7%higher than gray wolf optimization support vector regression.