| Require: Road segment features space: |
| Outputs: Road segment embedded features space: |
(1) | Define encoder parameters: |
(2) | Input layer: . |
(3) | Hidden layer1: , Activation = ReLu. |
(4) | Hidden layer2: , Activation = ReLu. |
(5) | Hidden layer3: , Activation = ReLu. |
(6) | Hidden layer4: , Activation = ReLu. |
(7) | Embedding layer , Activation = ReLu. |
(8) | Define decoder parameters: |
(9) | Hidden layer1: , Activation = ReLu. |
(10) | Hidden layer2: , Activation = ReLu. |
(11) | Hidden layer3: , Activation = ReLu. |
(12) | Hidden layer4: , Activation = ReLu. |
(13) | Output layer: , activation = Sigmoid. |
(14) | Define DAE model: model(encoder, decoder) |
(15) | fordo |
(16) | Fit input feature vectors to DAE model. |
(17) | Initialise weights randomly. |
(18) | Obtain reconstructed feature vectors . |
(19) | Compute the error difference: |
(20) | while error difference is not converging do |
(21) | Update weight parameters. |
(22) | end while |
(23) | Store weights parameters. |
(24) | Obtain the embedding features vector |
(25) | |
(26) | end for |
(27) | Return embedding space features |