Voids can be easily formed under the bottom plate of sluices under the effects of foundation consolidation settlement and long-term erosion by water flow, leading to engineering failure. The sluice bottom plates can be partially or completely hidden underwater for a long time. Generally, there are no conditions for drained testing, and drainage incurs time and cost, introducing difficulties into engineering safety testing. This paper proposes sensitive indices for detecting voids in underwater concrete structures based on the theory of mechanical elastic waves. In addition, a precise identification method for void defects based on the decision-tree algorithm is proposed. The technology for detecting voids in underwater concrete structures was applied to the concrete bottom plate of a sluice in Foshan, and its accuracy was verified through drilling at a sluice in Zhongshan., China. The results showed that the shock response intensity can be used as a sensitive index for calculating void area and height. A database of void defect samples was established based on 11 feature attributes, including shock response intensity, nine interval feature attributes selected from spectral features, and the location pattern of void defects. Through the pruning decision-tree classification algorithm, quantitative identification of void defects was achieved, with a verified accuracy of 93.33%. The loose and slight detachment areas in the box culvert of a sluice in Foshan were about 117.62 m2 and 4.03 m2, respectively, accounting for 14.05% and 0.48% of the total bottom plate area.