**2. Materials and Methods**

As shown in Figure 1, the algorithm of loop closure detection proposed in this paper includes the following key modules:

(1) The extraction of the semantic landmarks.

the following key modules:


**Figure 1.** Overview of the semantic loop closure detection algorithm based on topology graphs and convolutional neural network (CNN) features. **Figure 1.** Overview of the semantic loop closure detection algorithm based on topology graphs and convolutional neural network (CNN) features.

(1) The extraction of the semantic landmarks.

Among the key steps, steps (1)–(4) are algorithms for constructing semantic topological graphs, which are given in Section 2.1, and steps (5) and (6) are loop closure detection algorithms, which are provided in Section 2.2. Among the key steps, steps 1)–4) are algorithms for constructing semantic topological graphs, which are given in Section 2.1, and steps 5) and 6) are loop closure detection algorithms, which are provided in Section 2.2.

*Remote Sens.* **2020**, *12*, x 5 of 27

The method presented in this paper is different from previous methods as follows: The method presented in this paper is different from previous methods as follows:


(6) The calculation overall similarity for loop closure detection.
