Small target identification system and method for nuclear power underwater mobile shooting scene based on deep learning

The invention belongs to the technical field of nuclear power fuel management, and particularly relates to a small target recognition system and method for a nuclear power underwater mobile shooting scene based on deep learning. Comprising the following steps: S1, constructing a small target ROI; s2, identifying a small target; s3, segmenting and counting small target information; and S4, storing and displaying the small target. The method has the beneficial effects that compared with a direct identification method, a multi-model fusion method is adopted, and the accuracy of identifying the small target in the single image is greatly improved. Compared with single-time prediction, the prediction method adopting segmentation and statistics of multiple frames of images can effectively inhibit and identify the instability of small targets in a single image. 本发明属于核电燃料管理技术领域，具体涉及一种基于深度学习面向核电水下移动拍摄场景的小目标识别系统及方法。包括如下步骤：S1：构造小目标ROI；S2：识别小目标；S3：分割和统计小目标信息；S4：保存和显示小目标。本发明的有益效果在于：相对直接识别的方法，采用多模型融合的方法，大幅提高识别单张图像中的小目标的准确率