Predicción de información fotométrica espacial basada en red neuronal artificial de retropropagación

Near-field goniophotometric measurement, essential for obtaining photometric information, suffers from excessively prolonged measurement cycles and errors. To address these issues, an intelligent robotic near-field goniophotometric measurement system was independently developed. The associated algorithms were studied and verified. In the experiment, the luminance distribution characteristics of a planar light source were discretely measured at different spatial angles. Then, near-field photometric reconstruction and far-field inversion algorithms were studied, and their accuracy was compared with measurement standard values. Subsequently, a back propagation artificial neural network (BP-ANN) was employed to predict the spatial distribution of far-field photometric information for planar light sources at denser spatial poses, aiming to accurately fit the target spatial photometric distribution. The results showed that the average error of far-field luminous intensity inversion using near-field photometric measurement system was 2.6%, and the average predicted illuminance error of BP-ANN was less than 2.3%. The near-field photometric measurement system developed in this paper enables precise acquisition of spatial photometric information. The proposed spatial photometric distribution prediction method based on BP-ANN significantly enhances experimental measurement efficiency while reducing errors.