基于CNN-RF模型的广州地区土壤质地识别方法

Recognition Method of Soil Texture in Guangzhou Based on CNN-RF Model

不同土壤质地直接影响土壤水分渗透程度和农作物养分吸收,进而影响农作物的产量及质量,针对土壤质地难以开展高效、精准识别等问题,基于卷积神经网络-随机森林(CNN-RF)模型算法用于实现土壤质地高效、精准识别。首先用比重法测定土壤样本中砂粒、粉粒和黏粒的百分比,然后采用自主研制的便携图像采集装置,对广州地区的土壤进行1000个样本采集并对土壤研磨、筛选、拍摄,建立土壤样本质地和图像的数据库,提取图像中的颜色特征和纹理特征,利用CNN-RF模型并结合3种组合(颜色、纹理、颜色+纹理)方法对土壤样本中的黏粒、粉粒和砂粒百分含量进行回归预测。采用平均绝对误差(MAE)、均方根误差(RMSE)和判定系数(R^(2))进行模型回归性能评估。从混淆矩阵进行模型分类结果可知,预测砂粒的MAE、RMSE、R^(2)值分别为3.37、3.71和0.99;粉粒的MAE、RMSE、R^(2)值分别为3.48、3.79和0.98;黏粒的MAE、RMSE和R2值分别为3.38,3.76,0.99。与RF、KNN、VGG6-RF模型相比,这种CNN-RF模型得到的MAE值和RMSE值较小,R^(2)接近于1,其准确度为99.43%,因而性能更优。该方法具有简单、易用、快速、可靠和准确等优点,对岭南丘陵耕地土壤的优化管理和可持续利用具有重要意义。

Different soil texture directly affects the degree of soil water penetration and crop nutrient absorption,and then affects the yield and quality of crops.In view of the difficulty of efficient and accurate identification of soil texture,this paper uses convolutional neural network random forest(cnn-rf)model algorithm to achieve efficient and accurate identification of soil texture.The soil was ground,screened,and photographed to establish a database of soil texture and images.The color features and texture features in the images were extracted and analyzed by CNN-RF model combined with 3 combination methods(color,texture,color with texture).Mean absolute error(MAE),root mean square error(RMSE)and decision coefficient(R^(2))were used to evaluate the regression performance of the model.From the results of model classification using the confusion matrix,it can be found that the predicted MAE,RMSE,and R^(2) values of sand particles are 3.37,3.71 and0.99,respectively.The MAE,RMSE,and R^(2) values of silt particles are 3.48,3.79 and 0.98,respectively.The MAE,RMSE,and R^(2) values of clay particles are 3.38,3.76,and 0.99,respectively.Compared with RF,KNN,VGG6-RF models,the MAE value and RMSE value obtained by CNN-RF model are smaller.The R^(2) value is close to 1,and its accuracy is 99.43%.Therefore,the performance is better.This method has the advantages of simplicity,ease of use,rapidity,reliability,and accuracy,and is of great significance to the optimal management and sustainable utilization of cultivated land soil in hilly area of south of the Five Ridges.