SSD-based innovations for improved construction management

Amidst a global labor shortage, construction site workers face heightened pressure, with recruits needing swift adaptation, amplifying safety risks in complex environments. Yet, integrating image detection intelligence technology offers a solution, aiding managers in effectively tackling challenges and bolstering efficiency and safety in site management. This study addresses the challenge of accurately detecting objects in construction sites, where obstacles such as small objects, occlusion, and high object density hinder detection performance. Enhancing the Single Shot MultiBox Detector (SSD) model with tailored data preprocessing grounded in the DIKW hierarchy, known as Data Preprocessing's Fusion SSD (DPF-SSD), significantly improves detection accuracy. This novel DPF-SSD methodology can resolve incomplete or missing data, noise, and outliers. Firstly, it filters high-quality images that exhibit complexity, blurriness, and consistency with object characteristics. Secondly, it converts the image format and clarifies labeling rules for object characteristics. Thirdly, it loads corresponding label information and pixel coordinates for training alongside the object image. Finally, it integrates loss functions like L1 smooth loss and Softmax Loss to enhance feature clarity and improve identification accuracy. The effectiveness of the mentioned optimization was validated using a custom dataset in Python, implemented within the TensorFlow framework. Experimental results revealed that the enhanced SSD achieved an 81.97 % mean average precision (mAP), signifying a noteworthy 15.85 % enhancement compared to the original model. This advancement holds promise for enhancing productivity and safety within construction sites, thereby facilitating the automation of site management processes.