Focused on the issue that U-Net has a poor segmentation effect for small-volume pulmonary nodules, a segmentation method based on deep transfer learning was proposed, and Block Superimposed Fine-Tuning (BSFT) strategy was used to assist the segmentation of pulmonary nodules. Firstly, convolutional neural network was used to learn the feature information of large natural image datasets. Then, the learned features were transferred to the network for the segmentation of small pulmonary nodule image datasets. From the last sampling layer of the network, the network was released block by block and fine-tuned until the network completed the superimposition of the last layer. Finally, the similarity coefficient of Dice was quantitatively analyzed to determine the optimal segmentation network. The experimental results show that the Dice value of BSFT on LUNA16 pulmonary nodule open dataset reaches 0.917 9, which is obviously better than that of the mainstream pulmonary nodule segmentation algorithms.

In order to get better speech enhancement effect for hearing aids when used in the environment with non-stationary or multiple noise, which will lead a sharp decline effect of user experience, a Coherent Filter Generalized Sidelobe Canceller (CF-GSC) speech enhancement algorithm based on small size microphone array was proposed. Aiming at the weak correlation noise which caused by the waves, fans and other approximate white noise, as well as the strong correlation noise caused by the point or other competitive sources, coherent filtering and traditional Generalized Sidelobe Canceller (GSC) structure were utilized to remove weak correlation and strong correlation noise separately, the Voice Activity Detection (VAD) algorithm was also applied during this process. The simulation results show that the proposed algorithm can obtain enhancement effect by almost 2 dB compared with the improved coherent filter and traditional generalized sidelobe canceller method under the environment of a variety of noise, meanwhile, the speech intelligibility also gets obviously improved.

Biomolecular networks alignment is an important field, and it is an effective way to study biomolecular phenomenon. Adaptive Hungary Greedy Algorithm (AHGA) is one of the valid biomolecular networks alignment algorithms. Commonly, biomolecular networks have large scale and biological background, so the data of biomolecular networks are special. In order to get the alignment results of biomolecular networks in acceptable time, considering the biological significance when aligning them, two methods including MPI (Message Passing Interface) and CUDA (Compute Unified Device Architecture) were used to parallelize the adaptive hybrid algorithm. The methods were analyzed and compared to find the suitable one for biomolecular networks alignment.

The size of median filtering window influences the capability of filter and 3×3 filtering window can preserve image details. An adaptive median filtering method was proposed. The extreme value point in 3×3 filter window was taken as candidate noise. If the candidate noise is extreme value in 7×7 window, then it is a sure noise. If the median value of 3×3 filter window is not noise, then the central noise would be replaced with the median value. The steps above were repeated until there was no noise replaced. If there still existed large noise cluster, the noise would be replaced with the mean of the three signals near it. The experimental results show that the method can denoise impulse noise and protect the image details effectively.

This paper proposed a stable and precise algorithm for 3D mesh model. Firstly, the algorithm, based on the original topology of mesh model, realized quick location of intersectional area of mesh curve by combining the intersection test of the layers of nodes bounding box. Then the algorithm utilized the improved triangular intersection algorithm to calculate the discrete intersectional segments and re-triangulates every intersectional triangle. Through building the topology of intersectional segments and triangles, the algorithm could quickly trace and pick up the discrete segments, and classify and combine the local area, and realize the precise Boolean operation. The algorithm could effectively deal with all kinds of instances, and could be implemented in programs easily. And the experimental results prove that the algorithm accords with requirements of the project.