Although the existing autoregressive translation models based on recurrent neural network, convolutional neural network or Transformer have good translation performance, they have the problem of low translation speed due to low decoding parallelism. Therefore, a non-autoregressive model based learning rate optimization strategy was proposed. On the basis of the non-autoregressive sequence model based on iterative optimization, the learning rate adjustment method was changed, which means that warm up was replaced with liner annealing. Firstly, liner annealing was evaluated to be better than warm up; then liner annealing was applied to the non-autoregressive sequence model in order to obtain the optimal balance between translation quality and decoding speed; finally a comparison between this method and the method of autoregressive model was carried out. Experimental results show that compared with the autoregressive model Transformer, when the decoding speed is increased by 2.74 times, this method has the BiLingual Evaluation Understudy (BLEU) score value of translation quality of 41.31, which reached 95.34% of that of the Transformer. It can be seen that the non-autoregressive sequence model of liner annealing can effectively improve the decoding speed under the condition of reducing a little translation quality, which is suitable for the platforms with urgent need for translation speed.

Focused on the issue that lexicalized reordering models are faced with context independence and sparsity problems in machine translation, a reordering table reconstruction model based on semantic content for reordering orientation and probability prediction was proposed. Firstly, continuous distributed representation approach was employed to acquire the feature vectors of reordering rules. Secondly, Recurrent Neural Networks (RNN) were utilized to predict the reordering orientation and probability of each reordering rule that represented with dense vectors. Finally, the original reordering table was filtered and reconstructed with more reasonable reordering probability distribution for the purpose of improving the reordering information accuracy in reordering model as well as reducing the size of the reordering table to speed up subsequent decoding process. The experimental results show that the reordering table reconstruction model can provide BLEU point gains (+0.39) for Chinese to Uyghur machine translation task.

Concerning the defect that the traditional filtering window cannot be adaptively extended and the standard mean filter algorithm could blur edges easily, a new adaptive weighted mean filtering algorithm based on city block distance was proposed. First, the noise points can be detected with switch filtering ideas. Then, for each noise point, the window was extended according to the city block distance, and the window size was adaptively adjusted based on the number of signal points within the window. Last, the weighted mean of the signal points in the window was taken as the gray value of the noise points to achieve the effective recovery of the noise points. The experimental results show that the algorithm can effectively filter out salt-and-pepper noise, especially for the larger-noise-density image, and denoising effect is more significant.

For estimating and removing the salt-and-pepper noise point accurately in image, a new adaptive median filtering algorithm was proposed.Firstly, if the pixel in the center of n×n (n is an odd integer not less than three) template was the extreme value of all the pixels in the window, it was supposed to be probably a noise point. The pixel gray value in the sequence difference between the two scripts and a template sequence of the slope of the pixel gray value within the region were used to determine the mean quasi-adaptive noise point to be the real noise points. Finally, mean filtering was done on the noised pixels. Compared with median filter, the condition of detecting noises with this method has been largely enhanced. And the method can both effectively restrain noises and maintain details.

This paper integrated the characteristics of the grey absolute relation with the advantages of the median filter to combine the pixels within the n×n template into two sequences, where n is an odd number that is greater than or equal to 3. Then, the characteristics of the grey absolute relation were used to determine the similarity between the two sequences. Finally, the degree of similarity was adopted to determine whether the current pixel is noise or not, and then the value of median filter was used to replace the noise one. The experimental results show that this algorithm has better filtering effect than the standard median filter method and other filtering methods while keepings more details of the original image.