为了改善不同朝向、外形尺寸船舶目标检测效果，提出基于深层卷积神经网络的船舶遥感图像检测方法。采用拉普拉斯算子对原始船舶遥感图像作增强处理，采用中值滤波算法消除处理后的遥感图像所含噪声，将去噪后的船舶遥感图像输入到船舶检测模型中，引入可变形卷积的D-FPN网络获取疑似船舶候选区域图像，通过多尺度锚点设计满足船舶目标检测时的尺寸要求，将疑似船舶候选区域提取结果作为D-FCN网络的输入，完成旋转矩形框定位以及损失函数的设计，实现不同类型船舶目标的识别。结果表明：该方法可提升原始船舶遥感图像视觉效果，实现船舶目标检测；嵌入3个可变形卷积层并放置于ResNet网络后端，模型P、R、F1指标值最大、训练损失最低。
In order to improve the detection performance of ship targets with different orientations and dimensions, a ship remote sensing image detection method based on deep convolutional neural network is proposed. After sampling the Laplace operator to enhance the original ship remote sensing image, the median filtering algorithm is used to eliminate the noise contained in the processed remote sensing image. The denoised ship remote sensing image is input into the ship detection model, and the suspected ship candidate area image is obtained by introducing a deformable convolutional D-FPN network. The multi-scale anchor design meets the size requirements for ship target detection, The extraction results of suspected ship candidate regions are used as inputs to the D-FCN network, and after completing the positioning of the rotating rectangular box and the design of the loss function, the recognition of different types of ship targets is achieved. The experimental results show that this method can improve the visual effect of original ship remote sensing images; It can achieve ship target detection by embedding three deformable convolutional layers and placing them on the backend of the ResNet network. The model has the highest P, R and F1 index values and the lowest training loss.
2023,45(18): 174-177 收稿日期：2023-04-13
DOI：10.3404/j.issn.1672-7649.2023.18.032
分类号：B237
作者简介：李泳强(1997-),男,硕士研究生,研究方向为函数型数据分析
参考文献：
[1] 喻钧, 康秦瑀, 陈中伟, 等. 基于全卷积神经网络的遥感图像海面目标检测[J]. 弹箭与制导学报, 2020, 40(5): 15-19+23.
[2] 于野, 艾华, 贺小军, 等. A-FPN算法及其在遥感图像船舶检测中的应用[J]. 遥感学报, 2020, 24(2): 107-115.
[3] 周慧, 刘振宇, 陈澎. 利用改进特征金字塔模型的SAR图像多目标船舶检测[J]. 电讯技术, 2020, 60(8): 896-901.
[4] 周秦汉, 王振. 基于多尺度特征增强卷积神经网络遥感目标检测算法[J]. 电光与控制, 2022, 29(11): 74-81.
[5] 李森森, 吴清. 改进Mask R-CNN的遥感图像多目标检测与分割[J]. 计算机工程与应用, 2020, 56(14): 183-190.
