Visible Light Imaging Simulation And Intelligent Recognition Application For Targets On Sea Surface

With the rapid development of China’s economy,China’s shipping industry has become increasingly prosperous,with the port cargo volume,throughput and ship volume ranking among the top in the world.As an important tool for marine loading and transportation,ships play an important role in early warning,disaster relief,and transportation in the marine field.At the same time,the development of sea area emergency surveillance platform technologies such as towing umbrellas and low-altitude drones in recent years has provided new technical means for enhancing the target situational awareness of sea areas.Different from high-altitude platforms such as satellites,the low-altitude platform is equipped with a wide field of view photoelectric camera and is quickly deployed.It can realize large squint and multi-angle observations of moving targets such as ships in the sea.The application of the new detection platform technology to the artificial intelligence recognition of optical ship target is faced with such problems as the limited acquisition of real strabis-viewing multi-angle ship image data and the difficulty in designing the intelligent ship target recognition algorithm under the condition of scarce sample dataIn this paper,for the purpose of building a squint ship target simulation system based on the ocean background,some key technologies in the system are systematically studied.It has laid an important foundation for the practical and commercialization of the subsequent ship automatic management identification system.The main work and contributions of this article are as follows:(1)For the imaging simulation of the low-altitude squint visible light ship target,the geometrical and spectral characteristics of the ship target and the background of the ocean wave are comprehensively considered,combined with the oblique observation of the atmospheric transmission effect,the oblique optical imaging simulation modeling of the oblique target is carried out.The squint multi-angle imaging simulation of the ship target and the sea background in the visible light band is realized,which provides simulation data for the research of target recognition and other algorithms.(2)An improved recognition algorithm network structure is proposed.The improved network structure is based on Squeeze Net,and the shallow output and deep output stages of the network are combined and used as the final output to make the feature content richer.At the same time,the Adam optimizer is improved to reduce whether the Loss continues to reduce adaptation during model training.The learning rate must be adjusted to speed up the model convergence.Experiments show that the improved recognition algorithm proposed in this paper is superior to traditional algorithms in terms of recognition accuracy and operating efficiency.(3)In this paper,the actual data is insufficient to effectively train the typical ship target intelligent recognition algorithm,and the multi-angle simulation data of the typical ship target is introduced to expand the training sample.By using the training samples augmented by the simulation images,whether it is the traditional knn algorithm or the improved algorithm proposed in the paper,the recognition accuracy of typical ship targets has been significantly improved,which verifies the superiority of the algorithm recognition performance improvement by using the simulation images to expand the training samples.(4)In order to improve the target computing performance of typical ship,the paper adopts a parallel technology route based on GPU.Experiments show that the performance after parallel computing is improved by nearly 3 times compared with that before parallel.