基于残差网络特征融合的高光谱图像分类

韩彦岭, 崔鹏霞, 杨树瑚, 刘业锟, 王静, 张云

Classification of hyperspectral image based on feature fusion of residual network

HAN Yanling, CUI Pengxia, YANG Shuhu, LIU Yekun, WANG Jing, ZHANG Yun

表2 改进残差网络

Tab.2 Improved residual network

层数 输出尺寸大小 卷积核大小及个数 步长 激活层 conv1i 27×27 3 × 3 32 3 × 3 32 3 × 3 32 1×1 ReLU pool1 14×14 — 2×2 — conv2i 14×14 3 × 3 64 3 × 3 64 3 × 3 64 1×1 ReLU pool2 7×7 — 2×2 — Dconv1 18×18 5×5 64 1×1 ReLU Dconv2 9×9 3×3 64 1×1 ReLU M1 18×18 32 — — conv3i 18×18 3 × 3 32 3 × 3 32 3 × 3 32 1×1 ReLU pool1 9×9 — 2×2 ReLU conv4i 9×9 3 × 3 64 3 × 3 64 3 × 3 64 1×1 ReLU pool2 5×5 — 2×2 ReLU conv5 4×4 3×3 128 1×1 ReLU pool5 2×2 — 2×2 —