A Big Bang-Big Crunch Type-2 Fuzzy Logic System for Explainable Semantic Segmentation of Trees in Satellite Images using HSV Color Space

In recent years, new sensor technologies have increased the accessibility of high-resolution satellite images. The information in these images can help to improve activities like urban planning and growth analysis of cities. Additionally, information extracted from these images can be used for taking decisions related to infrastructure planning, e.g. identifying objects that might interfere with network assets like underground cables. To be able to justify the cost of network planning decisions a high degree of interpretability is required. Convolutional Neural Networks (CNNs) are the state of the art for segmenting these images, but like any black box model they do not offer any explanation for their output. In this paper, we present an approach on how to use a Fuzzy Logic System (FLS) for performing explainable semantic segmentation of trees in satellite images. The FLS uses the HSV (hue, saturation, value) of the pixels as inputs and was optimized by using an evolutionary algorithm called Big Bang Big Crunch. The best configuration for the Interval Type-2 FLS has an Intersection over Union metric measure of 60.6%, which is close to the results obtained from neural network, however the proposed FLS provides interpretable outputs which is highly needed for the real-world operation especially in the telecommunication domain.