| At present,there are various methods to measure soil moisture,but they all have some defects,such as the drying method has high accuracy of measurement results,but the measurement time is long and the timeliness is low,the sensor measurement method with short measurement time is high in timeliness,but it has the disadvantages of high cost of early deployment and later maintenance,only fixed-point measurement can be realized,and the deployment flexibility is low.The non-destructive and rapid measurement technology of soil moisture,which can meet the needs of scientific research and production accuracy,flexible deployment and low cost,has become a hot research direction.Image recognition technology based on deep learning is a fast developing method in recent years,which has mature applications in many fields,and has the characteristics of noncontact,high recognition rate and fast speed.However,at present,the application of deep learning technology in agriculture is far less extensive than that in industry.The main reason is that compared with industrial production,agricultural production has more uncertainties,more difficult to quantify and more difficult to obtain samples.In view of the above problems,this paper proposes a set of soil moisture classification application system based on deep learning technology,which solves the problems of few samples,many uncertain factors and difficult application when deep learning technology is applied to the rapid identification of soil moisture from the aspects of sampling,training,optimization and application.In the aspect of sample collection,the existing intelligent agricultural facilities are fully utilized to realize the expansion of the soil picture training set expansion system by expanding the function of the data collection gateway without increasing too much cost,and in particular,the training set samples can be directly verified by a temperature and humidity sensor to realize automatic supervised learning.It can quickly improve the training accuracy and realize the deployment and application of the model in actual production.In terms of algorithm implementation and optimization,this paper uses deep learning image recognition technology,uses self-built soil image data set,and divides the soil surface image into three categories according to the corresponding humidity through model training.The recognition accuracy of the training set is 95% or more,and the recognition accuracy of the test set is about 80%.Deep learning image recognition technology is feasible in the field of soil moisture measurement.Based on the technical characteristics,this method has the characteristics of non-contact,non-destructive,rapid measurement and relatively low cost.In the aspect of method application,aiming at the problems of information island and poor expansibility in the current agricultural information system,a four-layer system architecture based on the modularization of "collection,storage,analysis and application" is designed,which can easily access the data collected by the Internet of Things and quickly dock with external systems.Through the PC end and mobile app end,it is convenient to manage the functions of each model and apply data visualization. |
