Concerning the problems of the lack of a large number of relationships in the current Knowledge Graph (KG), and the lack of full consideration of the hidden information in the multi-hop path between two entities when performing relationship reasoning, a relationship reasoning method combining multi-hop relationship path information was proposed. Firstly, for the given candidate relationships and two entities, the convolution operation was used to encode the multi-hop relationship path connecting the two entities into a low-dimensional space and extract the information. Secondly, the Bidirectional Long Short-Term Memory (BiLSTM) network was used for modeling to generate the relationship path representation vector, and the attention mechanism was used to combine it with the candidate relationship representation vector. Finally, a multi-step reasoning method was used to find the relationship with the highest matching degree as the reasoning result and judge its precision. Compared with the current popular Path Ranking Algorithm (PRA), the neural network model named Path-RNN and reinforcement learning model named MINERVA, the proposed algorithm had the Mean Average Precision (MAP) increased by 1.96,8.6 and 1.6 percentage points respectively when using the large-scale knowledge graph dataset NELL995 for experiments. And when using the small-scale knowledge graph dataset Kinship for experiments, the proposed algorithm had the MAP improved by 21.3,13 and 12.1 percentage points respectively compared to PRA and MINERVA. The experimental results show that the proposed method can infer the relationship links between entities more accurately.

In order to solve the problem of imbalance of label distribution in an image dataset and improve the annotation performance of rare labels, a Multi Label Smoothing Unit (MLSU) based on label smoothing strategy was proposed. High-frequency labels in the dataset were automatically smoothed during training the network model, so that the network appropriately raised the output value of low-frequency labels, thus, the annotation performance of low-frequency labels was improved. Focusing on the problem that the number of images was insufficient in the dataset for image annotation, a Convolutional Neural Network (CNN) model based on transfer learning was proposed. Firstly, the deep convolutional neural network was pre-trained by using the large public image datasets on the Internet. Then, the target dataset was used to fine-tune the network parameters, and a Convolutional Neural Network model using Multi-Label Smoothing Unit (CNN-MLSU) was established. Experiments were carried out on the benchmark image annotation datasets Corel5K and the IAPR TC-12 respectively. The experimental results show that the average accuracy and average recall of the proposed method are 5 percentage points and 8 percentage points higher than those of the Convolutional Neural Network Regression (CNN-R) on the Corel5K dataset. And on the IAPR TC-12 dataset, the average recall of the proposed method has increased by 6 percentage points compared with the Two-Pass K-Nearest Neighbor (2P KNN_ML). The results show that the CNN-MLSU method based on transfer learning can effectively prevent the over-fitting of network and improve the annotation performance of low-frequency labels.

Aiming at the problems of the odor source localization algorithms by using multi-robot in indoor turbulent environment, such as the low utilization rate of historical concentration information and the lack of mechanism to adjust the global and local search, a multi-robot cooperative search algorithm combing Brain Storm Optimization (BSO) algorithm and upwind search was proposed. Firstly, the searched location of robot was initialized as an individual and the robot position was taken as the center for clustering, which effectively used the guiding role of historical information. Secondly, the upwind search was defined as an individual mutation operation to dynamically adjust the number of new individuals generated by the fusion of selected individuals in a class or two classes, which effectively adjusted the global and local search methods. Finally, the odor source was confirmed according to the two indexes of concentration and persistence. In the simulation experiments under two environments with and without obstacles, the proposed algorithm was compared with three kinds of swarm intelligent multi-robot odor source localization algorithms. The experimental results show that, the average search time of the proposed algorithm is reduced by more than 33% and the location accuracy is 100%. The proposed algorithm can effectively adjust the global and local search relations of robot, and locate the odor source quickly and accurately.

In the environment with unknown dynamic obstacle moving and target point, the radius of the repulsive force is often larger than the radius of the obstacle when the path is planned by the artificial potential field method, which leads to the collision of the dynamic obstacle with the robot. An improved dynamic path planning strategy of artificial potential field based on Morphine algorithm and non-completely waiting strategy was proposed. The non-completely waiting strategy was adopted when the dynamic obstacle collided with the robot on a side. The Morphine algorithm was used to localize the path when the dynamic obstacle collided with the robot face to face. Moreover, the rolling window theory was introduced to improve the accuracy of avoiding dynamic obstacles. Through the simulation tests, compared with the traditional artificial potential field, the proposed algorithm is shortened by 12 steps in the event of a side collision and 6 steps in the event of a face-to-face collision. Therefore, the improved algorithm is more effective in path smoothness and planning steps.

Aiming at the problems of the current self-localization algorithms for indoor mobile robot, such as the low positioning accuracy, increasing positioning error with time, the signal's multipath effect and non-line-of-sight effect, a new mobile robot self-localization method based on Monte Carlo Localization (MCL) was proposed. Firstly, through analyzing the mobile robot self-localization system based on Radio Frequency IDentification (RFID), the robot motion model was established. Secondly, through the analysis of the mobile robot positioning system based on Received Signal Strength Indicator (RSSI), the observation model was put forward. Finally, in order to improve the computing efficiency of particle filter, the particle culling strategy and particle weight strategy considering orientation of the particles were given, to enhance the positioning accuracy and the execution efficiency of the new positioning system. The position errors of the new algorithm were about 3 cm in both the X direction and the Y direction, while position error of the traditional localization algorithm in the X direction and the Y direction were both about 6 cm. Simulation results show that the new algorithm doubles the positioning accuracy, and has good robustness.