Analysis of Soil Type by Image Using Neural Networks

The study of soil and terrain is an important aspect in agriculture. Assessment of the soil condition both on already used lands and on new territories allows enterprises to assess the opportunities and risks of using these sites more accurately. The developed solution is based on terrain recognition using convolutional neural networks. It allows to classify terrain based on soil images, including aerial photography. The method classifies each pixel of the image using window traversal. As a result, it is possible to analyze each of the image sections in detail and consider the classes assigned to them, thus increasing the accuracy of recognition. The implemented neural network was tested on various sets of test data, where the results of terrain recognition accuracy of more than 90% were obtained. The method is supposed to be implemented in the process of soil and terrain assessment.