Fig. 1. Road extraction algorithm incorporating elemental multiplication and detail optimisation

Fig. 2. IEM block

Fig. 3. Comparison of receptive fields in different models. (a) IEM block sensory field; (b) IEM block without two Conv3×3 branches

Fig. 4. Depthwise convolution

Fig. 5. Pointwise convolution

Fig. 6. Classification of 2D planes

Fig. 7. Three-dimensional plane classification

Fig. 8. Comparison of heat maps. (a) Image; (b) Conv3×3; (c) Original IEM; (d) IEM

Fig. 9. Parameter-free attention mechanism of PFAAM

Fig. 10. Comparison of heat maps. (a) Image; (b) Direct output; (c) Output + RRN

Fig. 11. Output loss function value conversion curves

Fig. 12. Comparison of visualization results extracted by various methods

Fig. 13. Comparison of visualization results extracted by various methods