With the continuous development of the blockchain technology and its applications， the demand for interoperability among blockchains is increasing. However， the isolation and closeness of blockchain as well as the heterogeneity among different blockchains cause the "island of value" effect of blockchain， which seriously hinder the widespread implementation and sound development of blockchain based integrated applications. Cross-chain technology of blockchain solves the problems of data circulation， value transfer and business collaboration among different blockchains， and is also an important approach to improve the scalability and interoperability of blockchains. According to the degrees of the implementation complexity and the function richness of cross-chain technology， the cross-chain technology of blockchain was summarized and then classified into three types： the basic cross-chain mechanisms， the cross-chain protocols based on these mechanisms， and the cross-chain applications with system architectures. Finally， the existing problems in cross-chain interoperations were summed up， thereby providing systematical and theoretical reference for the further research on cross-chain technology of blockchain.

The gradient descent method has poor convergence and is sensitive to local minimum. Therefore, an improved NAG (Nesterov’s Accelerated Gradient) algorithm was proposed to replace the gradient descent algorithm in the Distance Regularized Level Set Evolution (DRLSE) model, so as to obtain a fast image segmentation algorithm based on NAG algorithm. First, the initial level set evolution equation was given. Second, the gradient was calculated by using the NAG algorithm. Finally, the level set function was updated continuously, avoiding the level set function falling into local minimum. Experimental results show that compared with the original algorithm in the DRLSE model, the proposed algorithm has the number of iterations reduced by about 30%, and the CPU running time reduced by more than 30%. The algorithm is simple to implement, and can be applied to segment the images with high real-time requirement such as infrared images and medical images .

To solve the low efficiency problem of data retrieval in indoor three-dimensional (3D) maps, an indoor 3D map data retrieval method based on octree was proposed. Firstly, the data was stored according to the octree segmentation method. Secondly, the data was encoded to facilitate addressing. Thirdly, the search data was filtered by adding a room interval constraint to the data. Finally, the indoor map data was retrieved. Compared with the search method without constraints, the search cost of the proposed method was reduced by 25 percentage points on average, and the search time was more stable. Therefore, the proposed method can significantly improve the application efficiency of indoor 3D map data.

For the fact that existing map updating methods are not good at map updating in indoor map environments, a hierarchical indoor map updating method was proposed. Firstly, the activity of indoor objects was taken as a parameter. Then, the division of hierarchy was performed to reduce the amount of updated data. Finally, a Convolutional Neural Network (CNN) was used to determine the attribution level of indoor data in network. The experimental results show that compared with the version update method, the update time of the proposed method is reduced by 27 percentage points, and the update time is gradually reduced compared with the incremental update method after the update item number is greater than 100. Compared with the incremental update method, the update package size of the proposed method is reduced by 6.2 percentage points, and its update package is always smaller than that of the version update method before the data item number is less than 200. Therefore, the proposed method can significantly improve the updating efficiency of indoor maps.

Delay Tolerant Network (DTN) has characteristics of dynamic topology changes and limited node storage space. A DTN Epidemic Routing with Congestion Control strategy (ERC²) method was proposed. The method was based on a Dynamic Storage State Model (DSSM). According to sensing network conditions, the threshold of node's semi-congested state was dynamically adjusted to reduce the possibility of network congestion by nodes. The ACK index and message management queue were added to make node storage state change randomly with network load, dynamically update and actively delete redundant packages. Single or mixed mode was selected for message forwarding according to different congestion states combining with advantages of Epidemic and Prophet routing, so as to achieve the purpose of preventing, avoiding and canceling congestion, realizing adaptive buffer management of nodes and dynamically controlling congestion of network. Simulations were conducted on the ONE(Opportunistic Networking Environment) platform using Working Day Movement (WDM) model. In the simulation, ERC² was 66.18% higher than Prophet in message delivery rate. The average latency of ERC² was decreased by 48.36%, and the forwarding number was increased by 22.83%. The simulation results show that ERC² has better network performance than Epidemic and Prophet routing algorithms in scenarios with different levels of congestion.

Concerning the problem of concept drift and noise in data stream, a new kind of incremental learning data stream ensemble classification algorithm was proposed. Firstly, a noise filtering mechanism was introduced to filter the noise. Then, a hypothesis testing method was introduced to detect the concept drift, and an incremental C4.5 decision tree was used as the base classifier to construct the weighted ensemble model. Finally, the incremental learning examples were realized, and the classification model was updated dynamically. The experimental results show that, the detection accuracy of the proposed ensemble classifier for concept drift reaches 95%-97%, and its noise immunity in data steam stays above 90%. The proposed algorithm has higher classification accuracy and better performance in the accuracy of detecting concept drift and noise immunity.

Focusing on the issue that the LBE (Local Background Enclosure) algorithm is over dependent on depth information and difficult to fully detect the object with complex structure, a RGB-D saliency detection algorithm based on the improved LBE features was proposed. Firstly, a set of segmentations was obtained by multi-level segmentation. Then, the depth saliency map was obtained by computing and merging the LBE features on each level segmentation map. Finally, a saliency map was obtained by adjusting the depth saliency map with color information and prior information. The experimental results show that compared with LBE algorithm, the precision of the proposed algorithm is slightly decreased and the recall is significantly improved, and the obtained saliency maps are much more close to the true values.

Splicing is the most universal image tampering operation, detection of which is effective for identifying image tamper. A blind splicing detection method was proposed. The method firstly analyzed the effects of different sub-bands on image splicing detection according to features of wavelet transform. High frequency sub-band was verified to be more appropriate for splicing detection both from theory analysis and experiment results. Secondly, the method conducted difference operation, rounded and made threshold to the coefficients as discrete Markov states, and calculated the state transition probabilities as splicing features. Finally, Support Vector Machine (SVM) was used as classifier, and the features were tested on Columbia image splicing detection evaluation datasets. The experimental results show that the proposed method performs better compared with other features and achieves a detection accuracy rate of 94.6% on the color dataset specially.

Concerning the low speed and low efficiency of learning in robot soccer when cooperating between multi-robots, a cooperative Q learning method based on the mental model of π calculus was proposed, in which the mental states were defined as the field state, goal, intention, action, cooperation, request, expanding knowledge，capability judging and connected intention, etc, and the combinational reward function was constructed. The validity of method was verified through experiments.