粗糙集在前视红外图像质量指标约简中的应用

摘要

To further reduce the attributes of forward looking infrared image quality preserved after the factor analysis, we abstract the accurate targets matching issue to the decision information system. Consequently, the attribute reduction method in rough set theory is adopted to remove the indices not significantly influencing the image quality, and we derive a reduct with five attributes. Since the attribute reduction in rough set theory theoretically prevents strong correlation between the attributes, therefore the theory is applied once again directly to reduce the initial fifteen attributes, and this reaches another reduct with five attributes. Contrast experiment reveals that the image quality assessment model of BP neural networks which uses the reduct of rough set either in conjunction with the factor analysis or just its own as the input has the advantages in both convergence rate and prediction accuracy.%为了对因子分析之后保留的前视红外图像质量指标作进一步约简,把目标是否正确匹配的决策问题抽象成决策信息系统,运用粗糙集理论中的属性约简方法去除了对图像质量影响不显著的指标,得到了包含5个指标的一种约简.由于粗糙集属性约简在原理上防止了属性之间的强相关性,所以直接采用粗糙集理论再次约简最初的15个指标,得到了包含5个指标的另一种约简.对比实验表明,无论是与因子分析结合使用,还是单独直接使用,以粗糙集约简结果为输入的BP神经网络图像质量评价模型在收敛速度和预测精度上都具有优势.