In order to further improve the speed and precision of multi-scale object detection, and to solve the situations such as miss detection, wrong detection and repeated detection caused by small object detection, an object detection algorithm based on improved You Only Look Once v3 (YOLOv3) was proposed to realize automatic detection of multi-scale object. Firstly, the network structure was improved in the feature extraction network, and the attention mechanism was introduced into the spatial dimensions of residual module to pay attention to small objects. Then, Dense Convulutional Network (DenseNet) was used to fully integrate shallow information of the network, and the depthwise separable convolution was used to replace the normal convolution of the backbone network, thereby reducing the number of model parameters and improving the detection speed. In the feature fusion network, the bidirectional fusion of the shallow and deep features was realized through the bidirectional feature pyramid structure, and the 3-scale prediction was changed to 4-scale prediction, which improved the learning ability of multi-scale features. In terms of loss function, Generalized Intersection over Union (GIoU) was selected as the loss function, so that the precision of identifying objects was increased, and the object miss rate was reduced. Experimental results show that on Pascal VOC datasets, the mean Average Precision (mAP) of the improved YOLOv3 algorithm is as high as 83.26%, which is 5.89 percentage points higher than that of the original YOLOv3 algorithm, and the detection speed of the improved algorithm reaches 22.0 frame/s. Compared with the original YOLOv3 algorithm on Common Objects in COntext (COCO) dataset, the improved algorithm has the mAP improved by 3.28 percentage points. At the same time, in multi-scale object detection, the mAP of the algorithm has been improved, which verifies the effectiveness of the object detection algorithm based on the improved YOLOv3.
To deal with the low resolution of depth images and blurring depth discontinuities, a depth image super-resolution method based on shape-adaptive non-local regression and non-local gradient regularization was proposed. To explore the correlation between non-local similar patches of depth image, a shape-adaptive non-local regression method was proposed. The shape-adaptive self-similarity patch was extracted for each pixel, and a similar pixel group for the target pixel was constructed according to its shape-adaptive patch. Then for each pixel in the similar pixel group, a non-local weight was obtained with the assistant of the high-resolution color image of the same scene, thereby constructing the non-local regression prior. To maintain the edge information of the depth image, the non-locality of the gradient of the depth image was explored. Different from the Total Variation (TV) regularization which assumed that all pixels obeyed Laplacian distribution with zero mean value, through non-local similarity of the depth image, the gradient mean value of specific pixel was estimated by non-local patches, and the gradient distribution of each pixel was fit by using the learned mean value. Experimental results show that compared with Edge Inconsistency Evaluation Model (EIEM) on Middlebury datasets, the proposed method decreases Mean Absolute Difference (MAD) of 41.1% and 40.8% respectively.
Many machine learning algorithms can cope well with prediction and classification, but these methods suffer from poor prediction accuracy and F1 score when they are used on medical datasets with small samples and large feature spaces. To improve the accuracy and F1 score of liver transplantation complication prediction, a prediction and classification method of liver transplantation complications based on Transfer Component Analysis (TCA) and Support Vector Machine (SVM) was proposed. In this method, TCA was used for mapping and dimension reduction of the feature space, and the source domain and the target domain were mapped to the same reproducing kernel Hilbert space, thereby achieving the adaptivity of edge distribution. The SVM was trained in the source domain after transferring, and the complications were predicted in the target domain after training. In the liver transplantation complication prediction experiments for complication Ⅰ, complication Ⅱ, complication Ⅲa, complication Ⅲb, and complication Ⅳ, compared with the traditional machine learning and Heterogeneous Domain Adaptation (HDA), the accuracy of the proposed method was improved by 7.8% to 42.8%, and the F1 score reached 85.0% to 99.0%, while the traditional machine learning and HDA had high accuracy but low recall due to the imbalance of positive and negative samples. Experimental results show that TCA combined with SVM can effectively improve the accuracy and F1 score of liver transplantation complication prediction.
In order to remove the effect of weather in degraded image, a fast haze removal algorithm for single image based on human visual characteristics was proposed. According to the luminance distribution of the hazy image and the human visual characteristics, the proposed method first applied luminance component to estimate coarse transmission map, then used a linear spatial filter to refine the transmission map and obtained the dehazed image by the atmospheric scattering model. Finally a new image enhancement fitting function was applied to enhance the luminance component of the dehazed image to make it more natural and clear. The experimental results show that the proposed algorithm effectively removes haze and is better than the existing algorithms in terms of contrast, information entropy and computing time.
Concerning the serious recession problems of the low-dose Computed Tomography (CT) reconstruction images, a low-dose CT reconstruction method of MLEM based on non-locality and variable exponent was presented. Considering the traditional anisotropic diffusion noise reduction is insufficient, variable exponent which could effectively compromise between heat conduction and anisotropic diffusion P-M models, and the similarity function which could detect the edge and details instead of gradient were applied to the traditional anisotropic diffusion, so as to achieve the desired effect. In each iteration, firstly, the basic MLEM algorithm was used to reconstruct the low-dose projection data. And then the diffusion function was improved by the non-local similarity measure, variable index and fuzzy mathematics theory, and the improved anisotropic diffusion was used to denoise the reconstructed image. Finally median filter was used to eliminate impulse noise points in the image. The experimental results show the proposed algorithm has a smaller numerical value than OS-PLS (Ordered Subsets-Penalized Least Squares), OS-PML-OSL (Ordered Subsets-Penalized Maximum Likelihood-One Step Late), and the algorithm based on the traditional PM, in the variance of Mean Absolute Error (MAE), and Normalized Mean Square Distance (NMSD), especially its Signal-to-Noise Ratio (SNR) is up to 10.52. This algorithm can effectively eliminate the bar of artifacts, and can keep image edges and details information better.