The paper evaluates two combined neural network-based models for predicting basic agrochemical soil parameters. A comparative study of the soil images obtained using three different optical devices was done related to the results of the model’s application. The images of the soil samples from different terrains were carried out under the same conditions with a camera—digital type; mobile phone type and document—camera. The obtained images and models with neural networks were processed in the Python language. Two combined digital models based on neural networks—VGG16 and ResNet50 are compared in the research. A comparative analysis and evaluation of the statistical average absolute error when using the optical devices was made. The digital camera and ResNet50 were found to have the lowest MAE of 13.76. The analysis of the literature review shows that there is a lack of combined models for the simultaneous prediction of basic qualitative soil agrochemical parameters. The comparative analysis of two such models presented in the article shows that they could find application in the simultaneous assessment of agrochemical soil parameters. The proposed digital models could be used in precise agriculture as a tool for agrochemical soil parameters assessment without time-consuming. The digital models could be realized as a software web-based application.