【摘要】 为促进计算机视觉技术的发展，提高遥感影像信息利用率，本文提出了复杂背景遥感影像敏感小目标细粒度智能识别。首先，利用中值滤波算法去除原始遥感影像中的干扰噪声，对图像实施灰度化处理后，通过背景分割阈值的合理选取，实现遥感影像内复杂背景与识别目标的划分，避免遥感影像背景信息对后续目标识别精度的影响。将处理后的遥感影像输入CNN网络，利用空间选择方法，在网络卷积层特征图输出的基础上，对遥感影像内的主体信息特征展开精炼提取，得到用于敏感小目标识别的遥感影像主体细粒度特征；结合ELM分类机制构建ELM小目标识别模型，实现遥感影像中敏感小目标的细粒度识别。实验证明，本文方法在面对包含多类别主体的遥感图像时，能做到特定小目标的精准识别，有效提高了遥感影像内信息的利用效率，对重要情报的获取具有重大意义。更多还原

【Abstract】 In order to promote the development of computer vision technology and improve the utilization of remote sensing image information, the proposed method proposes fine-grained intelligent recognition of sensitive small targets in complex background remote sensing images. The proposed method first utilizes a median filtering algorithm to remove interference noise from the original remote sensing image. After graying out the image, a reasonable selection of background segmentation threshold is used to achieve the division of complex backgrounds and recognition targets in the remote sensing image, avoiding the impact of background information in the remote sensing image on the accuracy of subsequent target recognition. Input the processed remote sensing image into the CNN network, and use spatial selection method to refine and extract the main information features in the remote sensing image based on the output of the network convolutional layer feature map, obtaining fine-grained features of the remote sensing image for sensitive small target recognition; Construct an ELM small target recognition model based on the ELM classification mechanism to achieve fine-grained recognition of sensitive small targets in remote sensing images.Experiments have shown that the proposed method can achieve precise recognition of specific small targets in remote sensing images containing multiple categories of subjects, effectively improving the utilization efficiency of information in remote sensing images, and has significant significance for obtaining important intelligence.更多还原