In underwater optical detection, due to the introduction of uneven light, the accuracy of underwater target recognition is reduced, especially when identifying underwater flexible targets. In this paper, a new type of depth wish-deformable convolutional neural network structure (DD-CNN) underwater target recognition algorithm is proposed for underwater flexible target recognition under the influence of non-uniform light. Firstly, the underwater invariant extraction algorithm based on non-subsampled contour let transform is used to extract the essential features of underwater images, and the underwater image after extraction is used as the input of the new depth wish-deformable convolutional neural network for target recognition. The experimental results show that the accuracy of using the proposed algorithm to identify underwater flexible targets is improved compared with the accuracy of the original structure, and the network convergence speed is faster.

在水下光学检测中，由于引入了不均匀的光线，特别是在识别水下柔性目标时，降低了水下目标识别的准确性。提出了一种新型的深度希望可变形卷积神经网络结构（DD-CNN）水下目标识别算法，用于在非均匀光影响下的水下柔性目标识别。首先，利用基于非下采样轮廓让变换的水下不变提取算法提取水下图像的本质特征，提取后的水下图像作为新的深度可变形卷积神经网络的输入，用于目标识别。 。