Table of contents

2023 7th International Conference on Artificial Intelligence, Automation and Control Technologies (AIACT 2023) is organized by Hong Kong Society of Mechanical Engineers. AIACT 2023 is a not-to-be-missed opportunity that distills the most current knowledge on a rapidly advancing discipline in one conference. Join key researchers and established professionals in the field of Artificial Intelligence, Automation and Control Technologies as they assess the current state-of-the-art and roadmap crucial areas for future research.

For a combination of reasons, most of the authors could not attend offline conference. AIACT 2023, which was planned to be held in Kunming, China from February 24 to 26, 2023, was changed to a virtual event and was held on February 24, 2023 through Tencent VooV software. This approach not only reduces people's travel, but also satisfies the need of communication. AIACT 2023 is targeted on providing opportunities to bring together the related researchers to share their most recent research achievements in these fields, and thus promote more advanced research.

More than 30 participants attended the meeting, most of them were from China. Three renowned speakers given speeches about their latest research and reports. They are: Prof. Dan Zhang, from York University, Canada; Prof. Wenqiang Zhang, from Fudan University, China; Prof. Hongliu Yu, from University of Shanghai for Science and Technology, China.

On the behalf of the conference organizing committee, I would like to express my great appreciation to the three keynote speakers (45 minutes each, including Q&A). In addition, the conference included one oral session and one poster session. In the oral session, each presentation was allotted 15 minutes. At the end of each session, the participants were engaged in discussions for future collaborations. A group photo was taken at the conference.

List of Committees are available in this pdf.

All papers published in this volume have been reviewed through processes administered by the Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

• Type of peer review: Double Anonymous

• Conference submission management system: Morressier

• Number of submissions received: 32

• Number of submissions sent for review: 26

• Number of submissions accepted: 16

• Acceptance Rate (Submissions Accepted / Submissions Received × 100): 50

• Average number of reviews per paper: 2

• Total number of reviewers involved: 20

• Contact person for queries:

Name: Coral Lu

Email: corallu@smehk.org

Affiliation: Hong Kong Society of Mechanical Engineers

While deep learning has achieved remarkable results for text classification, incremental learning for text classification is still a challenge. The main problem is that models suffer from catastrophic forgetting, which is they always forget knowledge learned before when labelled data comes sequentially and is trained in sequence. In this study, we propose methods of preventing catastrophic forgetting to handle unbalanced increased data. As an improvement over experience replay, our approaches improve the accuracy about 23.3% with 23% of all training data on Yahoo and 9.5% with 12% of all training data and on DBPedia.

Extreme multi-label text classification is a very important and challenging problem in this age of widespread internet access, with a wide range of application scenarios, such as web tagging, legal document annotation, commodity classification, etc. Most of the existing state-of-the-art methods are based on deep learning, however, there are still two problems: 1) the vector representation generated by most pre-trained models suffers from anisotropy and uneven distribution, which has a significant impact on the XMC task. 2) existing models are large in size and use many models for integration. Some of them even took hundreds of hours to train. This seriously affects the efficiency of the experiments. Therefore, in this paper, we propose CRAT-XML, which uses contrast adversarial learning to optimize text representation and enhance the acquisition of dependency relations between text and labels, thus reducing the need for integration at the representation level and achieving relatively high accuracy under low-resource and low-time conditions. Experimental results demonstrate that our model achieves SOTA results on a single model, while achieving a large reduction in training time and model size.

In landscape design, the text of landscape design plays an important role. However, in practice, designers are often faced with an industry situation where they need to complete a large number of design tasks in a short period. If language models are used to assist in the writing of design texts, it will be of great help not only to designers but also to the landscape design industry. A pre-trained language model trained on collected landscape design texts and used for landscape text-assisted authoring. It is better than the existing models in terms of perplexity, readability and manual evaluation. It has great potential to be used to improve the landscape design industry.

Collision-free navigation without pre-built maps is a promising and challenging problem. To make full use of the information in the environment, we propose a multi-input mapless navigation framework that combines both vision-based and ranging-based approaches. It takes the position and orientation of the robot relative to the target as the reference, incorporating RGB-D images and distance measurements as state inputs to the deep reinforcement learning model. In the framework, Dual-VAE is designed to encode RGB images and depth images simultaneously, in order to effectively mine the intrinsic information of RGB-D images. Moreover, we engineer a special reward function that enables the robot to navigate with a smooth trajectory and leave redundant space with obstacles for other tasks. Target-driven multi-input mapless navigation framework has been experimentally proven to achieve satisfactory performance in simulation environments. Our source code and simulation environments are released at https://github.com/YosephYu/MultiInput-Mapless-Navigation.

Deep reinforcement learning outperforms traditional methods in some domains. In this paper, we propose a novel reinforcement learning on policy (RL) algorithm, the Smoothing Clip Advantage Proximal Policy optimization Algorithm (SCAPPO), which extends the classical PPO algorithm where we exploit the smoothing properties of the sigmoid function to make full use of useful gradients. In addition, we provide more efficient gradients for policy networks effective gradients, aiming to solve the overfitting problem caused by the coupling of strategy and value functions. SCAPPO outperforms currently popular reinforcement learning algorithms in performance tasks in the Open AI Gym.

With the rapid development of electronic technology, the application of civil aircraft cabin electronics is becoming more and more advanced. The traditional cabin core network architecture can no longer meet current cabin application requirements. Meanwhile, there are multiple of devices with duplicated functions in cabin of civil aircraft, and the number of each device with duplicated functions is very large. In order to solve the problem of repetitive deployment of a large number of devices with duplicated functions in the cabin system, the civil aircraft cabin core network architecture is proposed to adopt a two-level daisy chain network topology which has many advantages such as high robustness, high scalability and less cables. In the design phase of civil aircraft cabin core network architecture, modeling and simulation of the network architecture through simulation tools is an important means to study, predict, justify and judge the network performance. In this paper, a two-level daisy chain network architecture model with about 67 nodes for civil aircraft cabin core network is developed and the network performance of the model mentioned above is simulated through OPNET simulation tool in order to perform a trade-off analysis and evaluate the cabin core network architecture.

The space inside the space capsule such as the space station sky and the core module is narrow, the size of the depth is large, and the bearing points are scattered and weak, there are some difficulties in loading, large size, different shape, complicated interface, scattered position, difficult position and pose adjustment, high assembly risk, etc.. To solve these problems, a six-DOF heavy-duty assembly robot is proposed in this paper. The robot adopts the combination of linear drive and rotating joint to increase the working space and bearing capacity of the robot. The kinematics and dynamics analysis model of the robot is established, and the path planning is carried out by means of quintic non-uniform B-spline interpolation method, considering fully the space constraints of the assembled equipment, a multi-objective trajectory optimization model based on NSGA2 algorithm is established to obtain the Pareto optimal solution set with time, impact and energy consumption as optimization objectives, prove that the equipment can be efficient and smooth. Reliable installation, with less impact and less power consumption under multi-objective optimization analysis.

The detection of dense small objects on the water surface is one of the hot topics in object detection. In this paper, a one-to-many label assignment strategy based on the OTA algorithm, which is applied to anchor-free detector, is proposed to improve the detection accuracy of dense small objects on water surface. To be specific, one-to-many label assignment means that a ground truth (GT) corresponding to multiple prediction boxes is conducted globally. The cost function used by OTA algorithm is improved to make the distribution of positive and negative samples more reasonable. Meanwhile, an efficient training strategy is designed to accelerate network convergence by adding L1 loss in the last stage of training. The results show that the proposed strategies achieve 87.9% average precision (AP). Compared to the original algorithm, we achieve 3.3% relative improvement for the detection precision of dense small objects.

This paper mainly discusses the planning method of shipboard transportation scheduling process on aircraft carrier from two aspects of deck moving path planning and motion coordination method for carrier-borne aircraft in deck. Two kinds of moving path planning algorithms between deck carrier aircraft stations based on standard motion pattern were proposed firstly. And the evaluation criteria of the planned moving paths were discussed. Then, basing on the network of deck moving paths generated by previous path planning algorithm, we developed a motion coordination method of deck carrier-borne aircraft basing on mixed integer programming formulations which can be used to solve the problems of coordination and collision in the transportation scheduling process of multiple carrier-borne aircraft effectively.

Strategic resources affect national development and security at all times. Actual utilization of strategic resources is crucial. In recent years, strategic resources such as coal and mineral powder usually need to be stored in closed stockyards after mining due to environmental protection, concealment, and other reasons. The existing stockpiling inventory system cannot autonomously complete the stockpiling inventory in the closed stockyard with low risk, low cost, and high precision. Therefore, this paper proposes a closed stockyard UAV intelligent inventory system. Our inventory system comprises a UAV system, a visual mapping system, and a classification and volume calculation system. In this paper, we focus on the navigation system belonging to the UAV system and propose a GNSS-denied environment navigation system based on lidar and ArUco markers. We conduct experiments in the dry coal shed of thermal power plants, one of the application scenarios. After analysis and comparison, it can be concluded that our intelligent UAV stockpiling inventory system can complete the task with high robustness and high precision with the assistance of our navigation system.

Suspension displacements and wheel center accelerations are important signals for suspension health monitoring systems to improve vehicle reliability and safety. The current way to obtain these signals is to install sensors on vehicles to conduct direct measurements. Usually, displacements are sampled at a slower rate than accelerations due to technical or economic limitations in real scenarios. This paper introduces a method for displacement reconstruction with low-sampling-rate displacement and high-sampling-rate acceleration measurements by formulating the reconstruction problem as a state estimation problem. A state-space model is established by identifying two data-driven models: a time-series Auto-Regressive model and a Finite Impulse Response model. Then, Kalman smoothing is used to estimate the displacement. A series of experiments have been done to show that the estimates from Kalman smoother coincide with the measurements.

The system composition and usage of UAVs are different from those of manned aircraft, and a large number of flight training sessions for UAV pilots can gradually make them understand and be familiar with the characteristics of UAV. In this paper, the quality evaluation indexes of UAV pilot training actions are sorted out and an evaluation model is constructed based on the Entropy Weight-TOPSIS method. Through application verification, the evaluation results of the model are consistent with the subjective evaluation results of flight instructors. The model can be applied to the evaluation of UAV pilot training actions quality, and at the same time has high evaluation efficiency.