AI Boosts Intelligent Upgrade in Communication Engineering Quality Inspection

Artificial intelligence (AI) is a scientific technology used to research, develop, and apply techniques for simulating, extending, and expanding human intelligence. The ultimate goal is to enable machines to possess human-like intelligence and continually evolve. With the ongoing evolution of computing capabilities and technologies, AI has been widely applied across various industries. For instance, the visual technology employing AI deep learning can achieve intelligent inspection and review, significantly enhancing inspection efficiency in scenarios such as vehicle and pedestrian detection on streets, facial recognition at railway stations and airports, and industrial quality inspection on production lines.

Challenges

For the delivery quality management of communication network projects, especially wireless base station projects, subcontractors typically take photos on-site with their mobile phones. Subsequently, the project quality team manually reviews these photos, resulting in overall low efficiency. Could AI deep learning serve as an alternative solution to manual auditing?

Yes, it can. However, when compared with the single scenario of facial recognition at railway stations and airports using fixed cameras, developing the intelligent auditing function at actual wireless base station delivery sites faces the following technical challenges:

• The quality of photos is uncontrollable, as images captured are easily affected by lighting conditions, hand tremors, motion blur, and other factors.
• The inconsistency in camera distance and angle affects the size and proportion of the objects.
• There are multiple quality checkpoints in one photo, and their distribution is unbalanced.
• The convergence of models exhibits a black-box nature, significantly impacting the accuracy of AI model recognition.
• Detection of communication equipment requires high-level expertise, while visual algorithms involve some level of abstraction, resulting in significant communication costs for technical and business teams.

Innovative Solution: Eliminating Pain Points

In response to these challenges, ZTE has been continuously innovating, providing a series of innovative solutions to facilitate intelligent quality inspection at communication engineering sites. Technically, leveraging deep learning-based target detection and instance segmentation neural network technologies based on multi-scale feature extraction, it can capture rich details of photographed objects under different shooting conditions, addressing the uncontrollability of mobile phone photography. ZTE also utilizes the class activation map (CAM) technology to analyze the discriminative degree of different image categories through CAM heat maps, adjusting model structure and loss functions to address the issue of uneven distribution of multiple inspection points. Drawing on practical experience, ZTE has developed the in-house EasyImage tool and established the AI ring R&D process (Fig. 1) to offer diverse solutions for incubating needs across various dimensions such as recognition accuracy, business relevance, and expertise, thereby significantly enhancing the automation of the R&D process.

• Providing out-of-the-box, sub-second response visual algorithms.
• Offering collaborative online annotation for multiple users, addressing the challenges of ambiguity and metric fuzziness in algorithm R&D through visualization of annotation results and model metrics.
• Leveraging a data lake to establish a closed loop for data and image samples within the service system, providing online analysis and reporting capabilities to aid in service decision-making and algorithm optimization.
• Integrating AI GPU with the iEPMS digital delivery platform to achieve end-to-end online data closed loop, replacing manual, simplistic, and repetitive tasks.

Intelligent Quality Inspection: Enhancing Quality and Efficiency

ZTE has completed the research and development of 17 types of quality inspection algorithms, equipped with capabilities such as multi-scene coverage, precise recognition, rapid response, and enhanced efficiency (Fig. 2). Leveraging the digital delivery platform iEPMS and scenario-based AI algorithms, ZTE binds AI models to inspection scenarios, using AI auditing to replace manual auditing. This enables instant inspection and auditing for quality work orders, greatly shortening the audit cycle. The average inspection period for work orders has been reduced from 4.5 days to 20 minutes, significantly cutting labor costs, minimizing the number of repeated site visits by partner personnel, and enhancing the efficiency of network construction. At the same time, it comprehensively ensures the quality of network construction, facilitating efficient network delivery.

The integration of AI technology into communication engineering scenarios marks a significant step toward automating high-frequency repetitive tasks, thereby replacing manual labor. This not only fosters innovation but also reshapes traditional industries and generates new operational models. Moreover, sharing AI models facilitates collaboration and knowledge exchange with customers, propelling technological development and industry progress.

In the future, ZTE will continue to explore the utilization of AR glasses, cloud rendering, spatial computing, and coordination capabilities to enable more innovative application scenarios such as AI inspection, AR acceptance, and AR training. Leveraging emerging intelligent technologies, ZTE aims to optimize business processes, facilitate the deep integration of digital technology with physical operations, collaborate with partners for intelligent and efficient ecosystem operations, deliver high-quality networks to customers, and achieve greater value realization.