由于水体对光的吸收和散射,水下光电图像具有低信噪比、低对比度等特点,导致目标难以识别,限制了水下光电成像装备的实际应用和发展。为提高目标的探测精度和识别率,提出包含一维并行卷积和子像素卷积的深度卷积神经网络,利用其从水下光电图像训练集中学习优化图像质量的参数,实现了去噪和对比度增强。实验结果表明,相比于经典去噪方法和对比度增强方法联合处理的结果,本文方法得到的峰值信噪比和均方根对比度分别平均提高了2.93 dB和14.41,能够有效地权衡去噪、对比度增强和亮度提升等,获得适合人眼视觉感受的图像,且处理单幅图像的平均速度是经典方法的9.46倍。利用测试集对网络进行测试,其在一定范围内较好地优化了图像质量,具有一定的泛化特性。

Underwater photoelectric images have a low signal-to-noise ratio and poor contrast because water absorbs and scatters light. This makes it difficult to identify targets and limits the practical applications and the development of underwater optoelectronic imaging equipment. To improve the detection accuracy and recognition rate of the target, we propose a deep convolutional neural network with one-dimensional parallel convolution and sub-pixel convolution. The convolutional neural network is used to learn the parameters that can improve the image quality from the underwater photoelectric image training set. Then, it can denoise and enhance the contrast for the test images. The peak signal-to-noise ratio obtained using our method showed an average improvement of 2.93 dB over the ratio obtained using the classic denoising and contrast enhancement methods; the root mean square contrast also increased by an average of 14.41. Therefore, our proposed method can effectively balance the denoising, contrast enhancement, and brightness enhancement. This will improve the image quality. The average processing speed of a single image is 9.46 times greater than that of the classic method. Finally, the network is tested using the test set. And our network could improve the image quality and provide a generalization characteristic within a certain range.