基于深度学习的遥感影像目标检测系统设计

Design of Remote Sensing Image Target Detection System Based on Deep Learning

摘要:

遥感影像目标检测虽然是一种极为有效的地表变化监测手段，但极易受到自然环境复杂性的影响，从而造成遥感影像中存在混合的杂质像素，导致目标检测准确性较差。为解决此问题，设计基于深度学习的遥感影像目标检测系统。建立深度学习框架，分层次连接遥感影像输入模块、图像帧预处理模块与目标检测算法模块，再借助影像目标输出结构单元，对已获得的遥感影像像素数据进行整合，实现系统硬件设计。在此基础上，提取遥感影像的多特征条件，完善现有的目标检测系统设计方案。通过分割多级目标节点的方式，得到遥感影像特征的小波分解结果，利用计算求得的边缘纹理系数，实现融合深度学习理论的遥感影像目标变化能力检测。实验结果表明，所设计遥感影像目标检测系统能够有效剔除杂质像素量，更能适应复杂多变的自然环境，获得更为准确的地表变化监测结果。

Abstract:

Although remote sensing image target detection is an extremely effective means of monitoring land surface changes, it is extremely susceptible to the complexity of the natural environment, resulting in mixed impurity pixels in remote sensing images, resulting in poor target detection accuracy. To solve this problem, a remote sensing image target detection system based on deep learning is designed. Establish a deep learning framework, connect the remote sensing image input module, image frame preprocessing module and target detection algorithm module at different levels, and then integrate the obtained remote sensing image pixel data with the help of the image target output structure unit to realize the system hardware design. On this basis, the multi-feature conditions of remote sensing images are extracted, and the existing target detection system design scheme is improved. By dividing multi-level target nodes, the wavelet decomposition results of remote sensing image characteristics are obtained, and the edge texture coefficients obtained by calculation are used to realize the detection of remote sensing image target change ability fused with deep learning theory. The experimental results show that the designed remote sensing image target detection system can effectively eliminate the amount of impurity pixels, can adapt to the complex and changeable natural environment, and obtain more accurate surface change monitoring results.