Research On Nondestructive Measurement Of Wheat Leaf Water Content Based On Machine Vision Technology

Wheat is a kind of grain crops widely cultivated,it is the important environmental factors for the leaf water content to reflect field water holding capacity and field sensation directly,and and directly influences the future wheat yield through irrigation,therefore,it’s an important index for the leaf water content to evaluate wheat yield.In the past,humidity sensors were used to measure soil moisture content,but it caused many inconveniences in farmland and irrigation operations,although some studies have started to solve those problems,there are still only in single leaf blade or the research algorithm is too old.This paper studies and designs the nondestructive testing model of wheat leaf water content based on machine vision technology.This water content prediction model uses machine vision technology to accurately,quickly and non-destructively predict the moisture content of wheat leaves.The image of wheat leaves was taken under natural lighting conditions,in order to improve the accuracy of the prediction model,firstly,enhanced processing of wheat leaf images,this part selected algorithms has histogram equalization,homomorphic filtering,and Retinex enhancement to eliminate the influence of natural light,and enhances the contrast between the image target and the background.In the image segmentation part,it used select the K-means cluster segmentation algorithm to segment the image because of the large gap between the image target color and the background color,and obtained a good segmentation effect.Extracted the color and texture features for the segmented image,and the color features have R,G,B,H,S,I,r,g,b,etc,the texture features have gray average,consistency,and entropy.Energy,correlation,moment of inertia,etc.Analyzed the correlation between these characteristic parameters and water content,select nine characteristic parameters with high correlation to build a BP neural network prediction model.Then PCA analysis about the 15 extracted feature parameters,and extracted the first 6 principal component features to build a PCA-BP neural network prediction model.Through analysis,the prediction accuracy of BP neural network and PCA-BP neural network prediction model is 95.9 % and 96.3 % respectively,it can be seen that after the PCA analysis the number of network input parameters decreases 1/3,the training time decreases,and the error is smaller.The two prediction models have high accuracy,so both of the prediction models can be applied to the actual prediction of wheat leaf water content.