Comparative Analysis of Methods for Cloud Segmentation in Infrared Images.

The increasing penetration of Photovoltaic (PV) systems in the power network makes the grid vulnerable to the projection of cloud shadows over PV systems. Real-time segmentation of clouds in infrared (IR) images is important to reduce the impact of noise in the short-term forecast of Global Solar Irradiance (GSI). This investigation presents a comparison between discriminate and generative models for cloud segmentation. Markov Random Fields (MRF), which add information from neighboring pixels to the prior, are included among the analyzed generative models. This investigation includes an evaluation of the performance of supervised and unsupervised learning methods in cloud segmentation. The discriminate models are solved in the primal formulation to make them feasible in real-time applications. The performances are compared using the j-statistic. Preprocessing of IR images to remove stationary artifacts increases the overall performances in all of the analyzed methods. The inclusion of features from neighboring pixels in the feature vectors leads to an improvement in the performances in some of the cases. The MRFs achieve the best performance in both unsupervised and supervised generative models. The discriminate models solved in the primal yield a dramatically lower computing time along with high performance in the segmentation. The performances of the generative models are comparable to those of the discriminate models when proper preprocessing is applied to the IR images.