Pruning unnecessary regions from weight kernels and filters used in convolutional nerual networks has proven to be effective for model compression. However, prior studies have often over-optimized for generic application areas, which are not ideal for specific problems in salient object detection. In this study, a novel pruning strategy denoted advantage rank (ARank) is developed for multiple salient object detection (Mul-SOD), in which salient priority criterion (SPC) is used to evaluate parameter contributions. The proposed SPC is a measurable standard for object priority that can be quantified and calculated using ARank. Simulation experiments demonstrate that Mul-SOD can be optimized by removing interference from low-SPC parameters. ARank effectively decreased interference and distinguished high-priority salient objects from multiple background objects.

事实证明，从卷积神经网络中使用的权重核和过滤器中删除不必要的区域对于模型压缩是有效的。但是，先前的研究通常针对通用应用领域进行了过度优化，这对于显着物体检测中的特定问题并不理想。在这项研究中，针对多显着物体检测（Mul-SOD）提出了一种新的修剪策略，称为优势等级（ARank），其中显着优先级标准（SPC）用于评估参数贡献。提议的SPC是对象优先级的可衡量标准，可以使用ARank进行量化和计算。仿真实验表明，可以通过消除来自低SPC参数的干扰来优化Mul-SOD。