Predictive Model Of Maize Moisture Stress During Jointing Stage Based On Multi-view Stereo Vision

Maize is one of the main food crops in China,and the soil moisture content is an important factor affecting maize yield during growth.Fast and accurate detection of the degree of maize moisture stress can timely guide water irrigation in the field and avoid reduced yields caused by maize plants due to water deficit.The problems of complicated operation,damage to plants and low accuracy in detecting moisture stress were focused.Based on the analysis and summary of domestic research results of moisture stress detection,a multi-view image acquisition system for maize was constructed,and took sequence images of maize jointing stage;Study the image pre-processing method,maize plant reconstruction method,multi-view point cloud data fusion method,maize point cloud skeleton extraction method,maize plant parameter automatic extraction method and built a maize soil moisture stress predictive model to non-contact measure maize plant type parameter and predict moisture stress.This work can provide a quick and convenient detection method for maize plant type parameters and moisture stress.The main research work and conclusions are as follows:(1)Multi-view maize image acquisition system was built and preprocessing maize images.Yefeng 203 maizes were taken as the research object and sequence images of 24 maize jointing stages were collected by Canon 70 D camera.16 images of every maize were taken,capture 3 perspectives at a time,a total of 1153 images were acquired.The obtained maize image information includes backgrounds such as flowerpots and ground.Using HSV to extract and divide maize plants and remove discrete points using morphological dilation and connected domain area,detection of maize plant edges using Scharr Filter and obtained the edge information of the maize plant.(2)There was missing area of point cloud model caused by occlusion of maize plant leaves.A corn reconstruction method based on multi-view stereo vision was proposed.Used SURF algorithm to detect and match feature points of pre-processed maize images to reconstruct maize single-view point cloud model.The iterative closest point(ICP)was used to merge the 2 maize cloud models data into the same coordinate system and the cloud skeleton was extracted by L1-median method.(3)Proposed an automatic measurement method of main plant type parameters based on geometric characteristics of maize plants.In order to realize the automatic measurement of three plant type parameters related to the degree of water stress in maize,including plant height,internode height,and full leaf length,a method for extracting plant type parameters based on geometric characteristics of maize plants was studied.The test results showed that the plant height,internode height,and full-leaf leaf length RMSE were 1.82 cm,0.42 cm,and 0.84 cm,respectively.It has high plant-type parameter determination accuracy and can meet the requirements of maize moisture stress prediction.(4)Construction and optimization of maize moisture stress model.A single-parameter maize moisture stress model based on internode height,full-spread leaf length,and plant height growth was established,and the soil moisture content was predicted by a linear regression model,an exponential regression model,and a polynomial regression model.The test results showed that the water stress prediction model based on internode height uses an index regression RMSE of the minimum of 2.9398%;the moisture stress prediction model based on plant height growth used a polynomial regression RMSE of the minimum of 2.9607%;water stress based on full leaf expansion and leaf length The prediction model uses polynomial regression with a minimum RMSE of 3.2000%.180 sets of maize plant type parameters and soil moisture content data were divided into 135 sets of training samples and 45 sets of validation groups.The LVQ neural network model,PNN neural network model,and ECOCSVM neural network model were used to construct maize moisture stress model.For the prediction model,45 samples of the verification group were used to verify the prediction accuracy of the model.The experimental results showed that the established PNN neural network maize moisture stress prediction model has the best prediction effect,the classification accuracy rate is 95.55%,and the prediction accuracy is high.