Underwater Image Enhancement Embedded Algorithm Based on Improved Ghost Module

Underwater image enhancement algorithms create conditions for subsequent processing tasks such as recognition, classification, and reconstruction. Usually, the stronger the performance of embedded devices, the higher their image processing capabilities and speed. However, due to limitations in application environment and current embedded development board performance, algorithms on embedded development boards should seek a subtle balance between high performance and high processing speed. This article is based on the FUnIE GAN network, using an improved Ghost module to compress the network model, using the Fused MBConv module to shorten the network training time, and using TensorRT strategy to further optimize the algorithm.