Improving and simulating urban landscape image recognition using combination optimization and fuzzy K-means algorithm

Abstract

Modern image recognition systems are pivotal in enhancing urban landscapes to support sustainable development and improving urban planning performance in a dynamic environment. Previous research focused on street-view panoramas is emerging as a new information source for urban studies due to the rapid advancements in image processing technology. However, challenges such as accuracy, feature extraction, uncertainty management, and a lack of approach integration remain unresolved. The research introduces a novel method combining a Combination Optimization (CO) strategy with a Fuzzy K-Means (FKM) clustering algorithm to address the challenges and achieve superior urban data analysis performance. CO specifically integrates the genetic algorithm (GA) to efficiently search for the optimal subset of features that maximize the performance of a convolutional neural network (CNN) based on extracted features. The Particle Swarm Optimization (PSO) aims to efficiently find the optimal feature subset by simulating the social behavior of particles, where each particle represents a feature combination to explore and exploit the solution space. The FKM allows for the clustering of mixed-use urban zones with greater accuracy, identifying complex patterns, and relationships that earlier research methods often overlook. It has proven highly effective in detecting and classifying mixed-use urban zones, delivering greater accuracy in recognition tasks than traditional clustering algorithms.