Mask RCNN在雾化背景下的船舶流量检测

基于掩码区域卷积神经网络（Mask region-convolutional neural network，Mask RCNN）模型检测海域卫星航拍图片中的船舶流量检测，实现雾化与模糊背景下的自动检测船舶数量与船只定位.基于搭建的Mask RCNN网络模型进行训练，依据输出的船只位置，与准确位置对比，不断调整模型参数提升准确度，再用训练好的模型参数对测试集中的图片进行检测.训练后进行测试的结果为:重叠度（intersection over union，IOU）取0.5时，边界框位置准确度达85.4%，船只数量检测准确度高达89.9%.上述结果表明，Mask RCNN网络模型可实现高精度的船舶流量监测. 

A Mask RCNN Based Algorithm for the Ships' Number and the Shape Detection

A ship detection algorithm based on Mask region-convolutional neural network (Mask RCNN) was proposed to detect both the position and the ship flow in the satellite image. The proposed algorithm was designed to automatically detect and locate the vessel positions in the observed sea area. Based on the data set generated, the Mask RCNN model was built and trained. According to the position and accuracy of the outputs, the model parameters were modified to further improve the detection accuracy. Then, the trained model was applied to testing the image of the data set, quantitatively evaluating the model. The testing results show that, when the intersection over union (IOU) is 0.5, the accuracy of boundary frame position can reach 85.4% and the accuracy of ship number detection can reach up to 89.9%. The simulation results reveal that Mask RCNN can be used to detect the ship flow precisely.