Aiming at the balance optimization problem of Lightweight Convolutional Neural Network （LCNN） in accuracy and complexity， an adaptive multi-scale feature channel grouping optimization algorithm based on fast Non-dominated Sorting Genetic Algorithm （NSGA-Ⅱ） was proposed to optimize the feature channel grouping structure of LCNN. Firstly， the complexity minimization and accuracy maximization of the feature fusion layer structure in LCNN were regarded as two optimization objectives， and the dual-objective function modeling and theoretical analysis were carried out. Then， a LCNN structure optimization framework based on NSGA-Ⅱ was designed， and an adaptive grouping layer based on NSGA-Ⅱ was added to deep convolution layer in original LCNN structure， thus constructing an Adaptive Multi-scale Feature Fusion Network based on NSGA2 （NSGA2-AMFFNetwork）. Experimental results on image classification datasets show that compared with the manually designed network structure M_blockNet_v1， NSGA2-AMFFNetwork has the average accuracy improved by 1.220 2 percentage points， and the running time decreased by 41.07%. This above indicates that the proposed optimization algorithm can balance the complexity and accuracy of LCNN， and also provide more options for network structure with balanced performance for ordinary users who lack domain knowledge.

Concerning the problem that the performance analysis method of Luby Transform (LT) codes under Gaussian elimination decoding algorithm is complicated and inaccurate, a novel performance analysis method based on probability transfer function was proposed. Firstly, for two LT codes with simple uniform degree distribution, the precise performance was studied and its quantitative expression was given. Secondly, the general LT code was investigated, and a simple but effective qualitative analysis method was proposed. Finally, the simulation work was done to verify the new method. In the comparison experiments with the traditional method which only gives the upper and the lower bounds of the rank of generated matrix, the maximum error of performance analysis results for simple uniform degree LT codes reduces to 0.0124, and the complexity of general LT codes decrease to O( k ²). Theoretical analysis shows that the proposed method can effectively guide the optimization design of LT codes in communication area.

To solve the problem of H.264 video copyright protection, a new blind watermarking algorithm was proposed. Based on the texture features of the picture, watermarking information was embedded on the Discrete Cosine Transform (DCT) domain of Instantaneous Decoding Refresh (IDR) frames. First, a rectangular sliding window was used to search the region of complex textures. Second, in the selected region, a 4×4 sub-block of maximum energy was chosen for embedding one watermarking bit. Last, one Alternating Current (AC) coefficient value of the selected 4×4 sub-block was modified adaptively. The experimental results show that Peak-Signal-to Noise Ratio (PSNR) decreases 0.15dB and the bitrate rises 0.49% on average, and the accuracy of watermark detection is above 91%; moreover, the algorithm can effectively resist the re-coded attacks of different Quantization Parameter (QP).