On-line Ship Target Detection In Optical Remote Sensing Images Under Cloud Interference Scene

Optical remote sensing image is widely used in target detection due to its high spatial resolution and rich imaging texture details.As a kind of important carrier of maritime transport,ships can be detected online by means of visible light remote sensing image ship target detection,which is of great significance in the field of high-timeliness tasks such as surface battlefield analysis and maritime rescue.However,ship target detection is easily affected by cloud,light,wavy waves and other factors,among which cloud cover and appearance interference on ship detection accuracy is particularly significant.In addition,the limited computing and storage capacity of the onboard computing platform also poses a great challenge to the online high-precision ship detection.In view of this,this paper conducts research on the online target detection method of ships with visible light remote sensing image under cloud interference,focusing on the breakthrough of high-precision detection method and online low-complexity detection method under cloud interference.Firstly,a lightweight ship target detection method based on sub-graph classification detection framework is proposed.On this basis,artificial design features are introduced to eliminate cloud false alarm to improve the adaptability to cloud interference scenes.The model is compressed by the model quantization method to further reduce the complexity of the method and improve the speed of online target detection.The specific research content of this paper is as follows:(1)In view of the current mainstream ship target detection methods,it is difficult to realize online detection due to the high computational complexity caused by the complex generation and detection models of a large number of pre-selected boxes,so a subgraph classification ship target detection method based on sparse MobileNetV2 is proposed.Subgraph cutting is used to replace the pre-selection box generation process,to avoid the huge amount of computation brought by pre-selection box generation algorithm,and through model pruning and layer number clipping method,the scale of MobileNetV2 model is compressed.Experimental results show that compared with other lightweight detection methods,this method not only guarantees certain detection accuracy,but also effectively reduces computational complexity and improves ship detection speed.(2)To solve the problem of high false alarm rate caused by cloud features being confused with ship features,an anti-cloud interference ship detection method based on Gray Level co-occurrence Matrix(GLCM)and Local Binary Pattern(LBP)features was proposed.GLCM and LBP features were used to further distinguish ship targets from cloud false alarm images,and SVM classifier was used to remove cloud false alarm.Experiments show that this method can greatly reduce the false alarm rate of ship detection in cloud interference images and improve the detection accuracy under the condition of increasing a small amount of computation.(3)The model quantization method is used to further reduce the size of the detection model for the fast and low complexity requirement of online detection task.Based on the analysis of the calculation mode of the detection method,the 32-bit floatingpoint number system used for model parameters is mapped to the 8-bit unsigned integer number system to achieve the model scale compression.Experiments show that this method can effectively reduce the storage capacity occupied by parameters and improve the speed of on-line ship detection at the expense of less detection accuracy.