Article Preview
TopIntroduction
In the current era of technological advancements, smart city management is swiftly emerging as a pivotal domain for achieving efficient urban operations and sustainable development (Wang, Chen, et al., 2023; Lou et al., 2023). Integrating advanced technology and data science into smart city management unleashes unprecedented potential, creating new opportunities to elevate the quality of public services and enhance the overall urban living experience. However, amid the rapid developments in this field, we inevitably confront a series of pressing issues (Mehmet & Aydin, 2023). Presently, smart urban management grapples with numerous challenges, including information silos, data fragmentation, and inadequate system integration, all of which are stark realities. The copious amount of data generated in the city is dispersed across different departments and systems, resulting in information isolation and a complex scenario for unified data management. This not only impedes a comprehensive understanding of urban management but also constrains the scientific and timely nature of decision-making. Effectively addressing these challenges requires the implementation of efficient means to foster collaboration across various facets of city management, enabling seamless data circulation (Al Mudawi et al., 2023; Saydirasulovich et al., 2023).
In the continuous evolution of smart city management, the use of object detection has become a crucial initiative to enhance urban management efficiency (Bai et al., 2023). The application of object detection technology enables city managers to identify and monitor various objects in the urban environment with greater precision and speed, optimizing resource allocation and enhancing the scientific and timely nature of decision-making (Terven & Cordova-Esparza, 2023). Through object detection, city managers can monitor key indicators such as traffic flow, environmental pollution, and the status of public facilities in real time, providing more accurate insight into various aspects of urban operations. This aids in swift issue resolution, prediction of potential risks, and the implementation of corresponding management measures. However, despite the significant potential of object detection in improving urban management efficiency, there are still some shortcomings in the current stage (Azizi et al., 2023). First, the accuracy of object detection systems in complex urban environments remains a challenge. Factors such as diverse lighting conditions, traffic situations, and crowd movement may lead to recognition errors in traditional object detection algorithms, impacting the system’s accurate understanding of the urban landscape. Second, current technology often faces computational and storage pressures when dealing with large-scale data. The vast size of cities requires the processing of a substantial amount of real-time data, which may result in slower system response times, limiting the effectiveness of object detection in real-time decision-making and issue resolution (Aboah et al., 2023).
Based on this, we propose YOLOv8_k, an enhancement to YOLOv8. Initially, we focused on optimizing the backbone network by introducing the BottleneckCSP module to replace the previous C2f module. This module, composed of the Bottleneck and CSP structures, facilitates the learning of residual features in the network and adjusts the depth and width of the feature maps. In comparison to the original C2f module, the BottleneckCSP module reduces memory consumption and alleviates computational bottlenecks. Additionally, the CBAM is incorporated to further enhance the model’s capability to capture critical information during feature extraction. Through CBAM, the model dynamically focuses on different channels and spatial positions of features, increasing sensitivity to target details and contextual information. This contributes to optimizing the representation of crucial features in object detection. In experimental evaluations, YOLOv8_k demonstrates superior performance and lower computational burden following targeted optimizations. By substituting the original C2f module with the BottleneckCSP module, we effectively reduce memory consumption and computational bottlenecks while maintaining model accuracy. This adaptation makes the model more suitable for scenarios in urban management where real-time responsiveness and efficiency are paramount.
The following are the three contributions of this article: