基于知识学习的多目标关联检测与识别方法

随着感知智能水平的日趋成熟,人工智能研究呈现出向认知智能发展的态势;作为人工智能在感知智能方面的代表领域,图像目标识别在向认知智能发展的道路上存在着人机之间以及机器之间知识难通用、可扩展性低等问题;为此,提出了基于知识的多目标关联判别框架,通过引入知识图谱,将目标特征知识进行语义化表达与规则化关联存储,构建了基于知识学习的多目标关联检测与识别方法,动态按需调用目标检测模型的同时,迭代式更新多目标关联的知识图谱,形成双向反馈学习循环;最后,通过相关仿真实验,验证了基于知识学习的多目标关联判别方法的可行性,为提高图像目标识别算法的知识通用化和可扩展性提供了新的思路.

A Method of Multi-target Association Detection and Recognition Based on Knowledge Learning

With the maturity of the level of perceptual intelligence, artificial intelligence research has shown a trend towards cognitive intelligence 1]. As a representative field of artificial intelligence in perceptual intelligence, image object recognition has problems in the development of cognitive intelligence between humans and machines and between machines, such as difficulty in common knowledge and low scalability. To this end, a knowledge-based multi-target association discrimination framework is proposed. By introducing a knowledge graph, the target feature knowledge is semantically expressed and regularized associative storage, and a multi-target association detection and recognition method based on knowledge learning is constructed, which is dynamically on-demand. While calling the target detection model, iteratively update the knowledge graph associated with multiple targets, forming a two-way feedback learning loop. Finally, through related simulation experiments, the feasibility of the multi-target association discrimination method based on knowledge learning is verified, and a new idea is provided for improving the knowledge generalization and scalability of the image target recognition algorithm.