When applying common object detection algorithms to detect small objects on high-resolution images, the down-sampling operation of the input images is inevitable due to the limitation of GPU memory. Accordingly, the details for characterizing small objects are lost. To resolve this contradiction, a small object detection method in a coarse-to-fine manner is presented. Specifically, some rough regions of interest (ROI) are firstly computed from low-resolution images. The prior knowledge of the positions of objects is used to guide the generation of ROIs. Then the features of small ROIs are recomputed from high-resolution images, and the features of large ROIs are obtained from the feature maps used to generate ROIs. The proposed method is validated on two datasets. One is a plant phenotyping dataset and the other is a public traffic sign dataset. Experimental results convincingly show the effectiveness of the proposed method.

Our deteriorating civil infrastructure faces the critical challenge of long-term structural health monitoring for damage detection and localization. In contrast to existing research that often separates the designs of wireless sensor networks and structural engineering algorithms, this paper proposes a cyber-physical codesign approach to structural health monitoring based on wireless sensor networks. Our approach closely integrates 1) flexibility-based damage localization methods that allow a tradeoff between the number of sensors and the resolution of damage localization, and 2) an energy-efficient, multilevel computing architecture specifically designed to leverage the multiresolution feature of the flexibility-based approach. The proposed approach has been implemented on the Intel Imote2 platform. Experiments on a simulated truss structure and a real full-scale truss structure demonstrate the system's efficacy in damage localization and energy efficiency.