基于树莓派的农田表土层土壤容重检测系统研究

Soil Bulk Density Detection System of Farmland Topsoil Based on Raspberry Pi

设计了一种基于树莓派的表层土壤容重检测系统,利用易于获取的土壤表面图像特征对表层土壤容重进行预测。提取图像的Tamura纹理特征以及图像的分形维数特征。经过验证,Tamura纹理特征中的粗糙度、对比度、方向度以及图像分形维数特征与土壤容重的相关性较高,相关系数分别为-0.754、-0.799、-0.806、-0.849,因而选用这4个参数作为预测模型输入。分别采用SVM回归模型和GRNN回归模型以及基于SVM、GRNN的Bagging集成模型对土壤容重进行预测。基于SVM、GRNN的Bagging集成模型预测结果同环刀法得到的结果进行相关性分析,决定系数R^(2)达到0.8641,预测结果的平均绝对误差(MAE)达到了0.0316 g/cm^(3),相对单一SVM回归模型和单一GRNN回归模型具有更好的预测结果。基于树莓派的农田表土层土壤容重检测系统的田间实时测量结果显示测量的平均绝对误差(MAE)为0.0412 g/cm^(3),满足了田间精准、快速检测的要求。

The soil bulk density of the topsoil layer is an important parameter of farmland soil,and it is of great significance to accurately measure and evaluate it.A vehicle-mounted surface soil bulk density detection system based on Raspberry Pi was designed.The system took soil surface images and predicted the surface soil bulk density using easily-obtained soil surface image features.Extracted the Tamura texture feature of the image and the fractal dimension feature of the image.After verification,the roughness,contrast,directionality,and fractal dimension features were highly correlated with soil bulk density,and the correlation coefficients were-0.754,-0.799,-0.806,and-0.849.So these four parameters were selected as the input of the prediction model.SVM regression model,GRNN regression model and Bagging integration model based on SVM and GRNN were used to predict soil bulk density.Based on the correlation analysis between the prediction results of the Bagging integration model of SVM and GRNN and the results obtained by the ring knife method,R^(2)reached 0.8641,and the average absolute error(MAE)of the prediction results reached 0.0316 g/cm^(3),and it had better prediction results than a single SVM regression model and a single GRNN regression model.The field test was carried out using the soil bulk density detection system of farmland topsoil based on Raspberry Pi.And the results showed that the average absolute error(MAE)of the measurement was 0.0412 g/cm^(3),which was in line with expectations and met the requirements of accurate and rapid detection.