Adaptive Fuzzy Neural Network for Efficient and Robust Image Segmentation on Edge Devices

Edge computing devices often face processing power, memory and energy resources constraints, and it is difficult to deploy traditional deep neural networks (DNN) that require complex and deep architectures. Researchers are studying the fusion of fuzzy logic systems (FLS) and neural networks to create lighter models that can deal with uncertainty more effectively[1]. Fuzzy logic is particularly good at handling fuzzy, noisy or inaccurate data, which is very suitable for edge environments where image quality may change or deteriorate.In this study, a method of superficially integrating adaptive fuzzy logic systems into neural networks (CNNs) to form adaptive fuzzy neural networks (AFNN) is proposed. By incorporating the adaptive fuzzy inference mechanism into the feature extraction layer, the model improves the ability to deal with uncertain features while maintaining a simple and lightweight structure[2]. In this way, AFNN can achieve high-precision segmentation without consuming too many resources. According to the experimental results, AFNN improves the detection accuracy of edges and small objects in segmented tasks, and shortens the complexity and inference time of the model. These results show the possibility of AFNN being practical in a resource-limited edge computing environment.