Deep residual network for soil nutrient assessment using optical sensors
C. T. Lincy, Fred A. Lenin, J. Jalbin- Plant Science
- Soil Science
Abstract
Background
Farmers need information regarding soil fertility at every location of their fields to attain a higher level of precision in nutrient management. Nonetheless, the acquisition and processing of soil samples are labor‐intensive and time‐utilizing, and the related cost remains high‐priced to farmers. Artificial intelligence is the most speedily growing area combined into approximately all aspects of human life. Soil macronutrients like nitrogen (N), phosphorous (P), and potassium (K) have a significant role in precision agriculture. There is a huge need for powerful and rapid measurement systems to measure accurately the macronutrients in the soil for optimal crop productivity, especially in site‐specific crop management system, where the application of fertilizer can be regulated spatially with respect to crop demand. Nevertheless, it can present a research direction to design an advanced scheme in order to predict the properties of soil. A portable sensor device is a basic need of an agriculture system for the accurate and rapid monitoring of soil macronutrients.
Aim
In this research, the soil nutrients identified from the collected soil samples using optical sensors are evaluated for their accuracy using a deep learning approach.
Methods
A deep residual network is exploited for the soil nutrient prediction after augmenting the gathered soil data. Finally, various performance evaluation measures, like mean squared error (MSE), mean absolute error (MAE), and root mean squared error (RMSE), are calculated to detect how accurately the sensor predicted the soil nutrients.
Results
From the experimental analysis, it is stated that the proposed model attained low MSE value of 4.59 e−09, the low RMSE value of 6.78 e−05, and the low MAE value of 4.66 e−05 for N prediction. Likewise, the proposed model attained the least MSE value of 1.41 e−05, the least RMSE value of 0.0003, and the least MAE value of 0.0001 for P prediction.
Conclusion
Finally, for K prediction, the proposed model achieved the least MSE value of 1.54 e−06, least RMSE value of 1.24 e−03, and the least MAE value of 1.38 e−05.