AXHU’s contribution to world-class research of AI
International Office, Anhui Xinhua University
Dec 15, 2022
Recently, by AXHU’s Prof. Wang Qijin-led academic team, a high-level research outcome titled “M2YOLOF: Based on Effective Receptive Fields and Multiple-in-Single-out Encoder for Object Detection” (https://doi.org/10.1016/j.eswa.2022.118928) is published on the “Expert Systems With Applications,” a top AI journal of the Chinese Academy of Sciences (CAS), which is also one of the first-class publications recognized by the Chinese Association for Artificial Intelligence (CAAI). In 2022, its IF (Impact Factor) indicates 8.665 and JCR (Journal Citation Reports) is of “CAS-based Q1 Divide,” and the H-index (of Google Scholar) ranks fifth in the category of AI Research. Specifically, this academic breakthrough in the scope of AI research notches a benchmark, and boosts AXHU’s scientific advance and discipline building in all respects.
This academic paper conducts a challenging research on the trade-off between Efficiency and Precision of the real-time object detection task. As for the traditional object detector, to increase its precision/accuracy is subjected to a huger time consumption, which is virtually impracticable happening in the real situations and conditions. Therefore, the scientific researchers put forward M2YOLOF, i.e., an object detection algorithm, to construct the multiple-in-single-out encoder to strengthen the local feature and global representation of each multi-scale object. The self-focusing concept is embedded into M2YOLOF to capture the large-scale object based on Visual Recognition Challenge, and exploit Anchor mechanism to generate training samples. Meanwhile, the research fellows can take advantage of “effective receptive fields” to design the Selection Policy for dynamic samples, and to rationalize the quantity of positive samples. The experiments testify that this approach enjoys obvious merits regarding precise detection as well as speed of reasoning and calculating.
Previously, Prof. Wang along with his academic fellows has joint-authored the paper—Enhancing representation learning by exploiting effective receptive fields for object detection—, and contributed it to the international-leading AI journal, “Neurocomputing,” issued by the CAS (its IF indicates 5.779 and JCR is placed in the “CAS-based Q1 Divide”). Remarkably, the M2YOLOF-driven solution will push the boundaries of AI research relating to the effective receptive fields for object detection.