Figure 1 | Scientific Reports

Figure 1

From: Convolutional Neural Networks enable efficient, accurate and fine-grained segmentation of plant species and communities from high-resolution UAV imagery

Figure 1

Scheme showing the CNN-based segmentation implemented using the U-net architecture. The input tile (left), an extract of 128 × 128 pixels, is input for multiple convolutions (encoding layers), which extract decisive spatial patterns at the respective scale (grey). Subsequently, the feature maps are resampled (max pooling) to a coarser resolution before the next convolution process is being applied (etc.). Eventually, the spatial features learned at the different spatial scales are combined through a series of decoding layers (up-convolutions) to segment the spatial extent of the target class (right, pink).

Back to article page