Experimental Study on Convolutional Neural Network-based Soil Moisture Feature Recognition

Soil moisture is one of the important physical characteristics of soil, and it is the basis for judging whether the soil needs irrigation. Based on the theory of deep learning and the technology of convolution neural network, this study realizes image-based soil moisture recognition, thus realizing the automation of soil irrigation. The convolution core, convolution layer, pool layer and the corresponding parameter feedback mechanism are constructed according to the characteristics of soil images with different humidity, based on which, the convolution neural network for soil moisture identification is constructed. Trained by 1200 soil image samples with corresponding humidity, a stable deep learning network has been established. Experiments show that the accuracy rate of soil moisture identification of the deep learning network reaches 85%, reaching the practical level.