Significant success has been achieved in human pose estimation networks based on heatmap detection. However， the methods based on heatmap detection has a large number of parameters due to redundant computations， quantization errors， and the requirement of heatmap decoding. To address these issues， a Lightweight pose estimation Network based on Non-globally dependent Integral Regression （Lite-NIRNet） was designed to reduce redundant computations in the network by employing Partial Convolution （PConv）， which made the network more lightweight. To respond to the information loss caused by PConv， a Coordinate Attention （CA） mechanism was introduced to fuse inter-channel features， thereby enhancing the network performance. Additionally， a Non-globally dependent Integral Regression （NIR） module was designed to incorporate coordinate supervision to the network， which reduced the influence of quantization errors on network performance. The proposed NIR was able to reduce the bias produced by traditional integral regression during expectation calculations effectively， balancing better learning gradients with lower bias. Experimental results show that compared with the advanced High-Resolution Network （HRNet）， Lite-NIRNet reduces the number of parameters and computational complexity by 73.0% and 63.4%， respectively， on COCO validation set， and achieves the mean Average Precision （mAP） of 72.8% without additional heatmap decoding. Furthermore， on MPII validation set， Lite-NIRNet can also achieve a good balance between network performance and complexity.

Interstitial Lung Disease （ILD） segmentation labels are highly costly， leading to small sample sizes in existing datasets and resulting in poor performance of trained models. To address this issue， a segmentation algorithm for ILD based on multi-task learning was proposed. Firstly， a multi-task segmentation model was constructed based on U-Net. Then， the generated lung segmentation labels were used as auxiliary task labels for multi-task learning. Finally， a method of dynamically weighting the multi-task loss functions was used to balance the losses of the primary task and the secondary task. Experimental results on a self-built ILD dataset show that the Dice Similarity Coefficient （DSC） of the multi-task segmentation model reaches 82.61%， which is 2.26 percentage points higher than that of U-Net. The experimental results demonstrate that the proposed algorithm can improve the segmentation performance of ILD and can assist clinical doctors in ILD diagnosis.

The accurate estimation of the Point Spread Function (PSF) is the key point in image restoration. For the unknown PSF parameter of defocus blur, an estimation method was proposed based on blur spectrum characteristic of image edge. Specifically, the blur spectrum feature of basic edge was analyzed, and then the edge model of natural image was treated as reference image. Furthermore, the max spectrum similarity was analyzed to obtain the right parameter between the image to be restored and the blurred reference image with defocus parameter in a continuous range. The experimental results show that the proposed algorithm suits large scale defocus blur images and has strong anti-noise ability.

Fast and accurate Point Spread Function (PSF) estimation method is the premise to obtain good results on the blur image restoration. To solve the deficiency of the discrete PSF of defocus-blurred and motion-blurred images, a discretization method was proposed based on the combination of geometric property of degradation model and sub-pixel estimation. Specifically, the principle of weight allocation was defined, which was related to the distance with the neighboring pixels. Thus, the discretization of PSF was realized. Finally, the experimental results illustrate that the proposed method improves result precision and outperforms the traditional one on visual quality, sharpness evaluation function, Peak Signal-to-Noise Ratio (PSNR) and Improved Signal-to-Noise Ratio (ISNR).

Current three-dimensional reconstruction algorithms of the computer vision field have limitations that they need to deploy and calibrate the cameras around the scene, or they need a structure light. Furthermore, these algorithms are not robust enough to every object. A new kind of four camera array reconstruction algorithms which properly combined the image registration algorithm and the camera array method was proposed to solve the robustness and limitation problems. It does not need calibration or structure light support. The experiments based on complex indoor sense with shadows demonstrate that this method is able to do dense point cloud reconstruction robustly and can overcome the shortcomings of current reconstruction algorithms.

To address the issue of insufficient intelligent analysis and autonomous decision-making capabilities as well as low fault diagnosis efficiency in building equipment operation and maintenance processes， a novel fault diagnosis method based on knowledge graphs and multi-task learning was proposed. Firstly， an operation and maintenance-oriented knowledge graph was constructed， and multi-source heterogeneous data were extracted from building equipment systems by using natural language processing and entity linking techniques， thereby obtaining rich knowledge representation. Secondly， in the case of few-shot labeling， multi-source symptom associated identification was explored. And unlabeled data were used to optimize the model parameters iteratively through self-training and co-training strategies， so as to improve generalization capability of the model. Finally， when designing fault root cause localization technology based on deep knowledge reasoning， a probabilistic graphical model was utilized to trace the fault propagation paths in complex equipment systems， thereby enhancing the accuracy and interpretability of fault analysis. Simultaneously， fusion mechanism was introduced into multi-task learning framework， thereby improving performance of the proposed method on fault diagnosis tasks. Experimental results demonstrate that the proposed method achieves a fault diagnosis accuracy of 92% with an average diagnosis time of 6.5 seconds per case， outperforming the comparison models in evaluation metrics such as accuracy， precision， and recall.