Research On Soil Moisture Content Based On Tomato Leaf Image

The research on crop based on computer vision image technology mostly focuses on the prediction of leaf water content and chlorophyll,while the research on soil water content based on leaf image is less.Therefore,this study attempts to use the visual image of tomato leaves to predict the soil moisture content.Based on the extracted characteristic of tomato leaves,the prediction model of water content of soil)is established to provide the basis for reasonable irrigation of greenhouse crops,so as to achieve water saving and improve the yield of tomatos more efficiently.There are four main research contents:(1)The tomato leaves were pretreated by normalization,filtering,graying,segmentation,morphological operation and restoration.Specially,image filtering and segmentation are the key points in the preprocessing process.Through the comparative analysis of four filtering methods and three segmentation algorithms,it is found that Gaussian filtering method has the best filtering effect on tomato leaf image,and the improved PSO+OTSU segmentation algorithm has the least iterations,the shortest running time,and the algorithm has the best segmentation effect on tomato leaf image.(2)Forty four features and parameter combinations of tomato leaf color,shape and texture were extracted.Based on Pearson correlation coefficient method,features with relatively high correlation with water content of soil were selected to construct prediction model.(3)Wavelet,RBF and GRNN neural network were used to establish the prediction model of water content of soil based on 60 groups of data collected before and after two times and the data removed.Through comparison,it was found that GRNN neural network had the best prediction effect whether the data were removed or not,and the prediction accuracy of the model after removing invalid data was higher.If don’t removed data,the model’s R2 is 0.31114,the RMSE is 1.2227,and the MAE is 0.98246;if removed data,the R2,RMSE and MAE were 0.89855,0.40884 and 0.37691,respectively.When the water content of soil reaches a certain level,it is feasible to predict the water content of soil based on tomato leaf image.(4)Using MATLAB APP Designer design tool,the computer-based soil moisture prediction software is designed,which can read the leaf image,preprocess the leaf and predict the soil moisture.