A Self-Adaptive CPS-Based Object Recognition Framework for Smart Glasses Using Dist-YOLOv3-Xception with Attention
Abstrak
The Industrial Era 4.0 has emerged as a response to the changes occurring in the world in a dynamic, unexpected, and uncertain manner. This situation requires analytical, predictive, and adaptive capabilities in an intelligent environment. This affects real-world object recognition in vision systems, which are frequently limited to specific signals. Thereby, it creates an adaptive gap. One potential solution to this problem is the development of self-adaptive cyber-physical systems (hereafter, SACPS) to enhance adaptability in recognizing diverse real-world objects. This paper introduces the SACPS model through an extended machine learning/deep learning model applied to smart glasses, which can detect and calculate object distances adaptively. The components of the developed model comprise smart glasses, contextual knowledge, and adaptive requirements based on the SACPS concept. We developed a pre-trained model by combining the Dist-YOLOv3 algorithm with Xception and an attention layer to obtain more optimal results. This research compared the new pre-trained model with those from previous research. Based on the evaluation, the model demonstrates improved performance compared to the baseline when tested on the KITTI dataset, recording a mean Recall (mRec) of 45.21%, mean Precision (mPrec) of 14.73%, and mean Average Precision (mAP) of 30.04%. Additionally, the adaptive system's response to increasing light intensity below 50 revealed good stability, with average post-enhancement brightness reaching 100.0703 (pixel intensity scale). These results demonstrate the significant potential of our model in handling changing environments with strong adaptation in diverse real-world object recognition scenarios. In the case of smart glass, the employment of SACPS can provide good adaptability in predicting distance and increasing light intensity.
Identifier
- DOI: https://doi.org/10.31449/inf.v49i15.8853
- OpenAlex: https://openalex.org/W4416609035