Using an improved back propagation neural network to study spatial distribution of sunshine illumination from sensor network data

The study of light distribution in orchards is very important for enhancing agricultural production. Nonlinear massive data, amounting to more than 190 MB, were collected over a 6-month period. Information such as the location, illumination, and time was obtained from wireless sensor networks, while that of canopy density and slope aspect was obtained through manual surveys. This paper proposes an improved back propagation (BP) neural network to study sunshine illumination distribution by exploiting these data. The basic BP neural network is divided into Q groups, each of which receives R samples and is trained individually using a gradient descent algorithm. Every grouped neural network records its error at the end of each training round. The new weights and thresholds, selected according to these error values, are employed in the next round of training, and the training process does not terminate until the error is within the desired goal. Finally, to verify the validity of the algorithm according to various criteria, the improved BP neural network is used to study sunshine illumination in an orchard. Our experiments show that the improved BP neural network algorithm performs better than traditional algorithms including the spline interpolation, Kriging, and basic neural network algorithms, and yields an accurate sunshine illumination distribution that can be used to improve agricultural production.