An effective deep learning model for grading abnormalities in retinal fundus images using variational auto-encoders

Diabetic retinopathy (DR) and Diabetic Macular Edema (DME) are severe diseases that affect the eyes due to damage in blood vessels. Computer-aided automated grading will help clinicians conduct disease diagnoses at ease. Experiments of automated image processing with deep learning techniques using CNN produce promising results, especially in the medical imaging domain. However, the disease grading tasks in retinal images using CNN struggle to retain high-quality information at the output. A novel deep learning model based on variational auto-encoder to grade DR and DME abnormalities in retinal images is proposed. The objective of the proposed model is to extract the most relevant retinal image features efficiently. It focuses on addressing less relevant candidate region generation and translational invariance present in images. The experiments are conducted in IDRID dataset and evaluated using accuracy, U-kappa, sensitivity, specificity and precision metrics. The results outperform compared with other state-of-art techniques.     

Vortex analysis and flow field optimization for the particle classifier with three rotary cages

The particle classifiers with three rotary cages have significant advantage in powder handling capacity. The flow field inside the classifiers were investigated from the perspective of vortex using Q criterion. Formation process and distribution of the vortices were intuitively exhibited. Structure of the guide cone was further optimized, and the classifying performance of the classifiers was evaluated. The results show that complex vortex structures were formed inside the classifier. The large-scale columnar vortex under the guide cone oscillates irregularly. This unwanted vortex is eliminated by extending the guide cone. The structure of the guide cone has little effect on the cut size, but the optimized guide cone with the long cylinder and cone significantly enhances the separation degree of the fine and coarse particles. The classifier obtains finer silica powder with a median size of 2.5 μm and higher Newton efficiency about 71.5%.     

案例推理的故障诊断技术研究综述

基于案例推理(case-based reasoning,CBR)的故障诊断作为一种新的智能诊断技术,模拟人类求解问题的思路,通过历史案例发现新问题的解。概述了CBR的理论基础和基本原理;从工作过程和集成框架两个方面综述了CBR故障诊断技术的研究现状,其中工作过程包括案例的表示、检索和重用,以及案例库的维护,集成框架包括CBR与基于规则推理、CBR与人工神经网络以及CBR与多智能体等三种情况;比较了六种故障诊断技术的特点及应用范围;总结了CBR故障诊断技术有待解决的问题。     

Skin detection for single images using dynamic skin color modeling

Up-to-date skin detection techniques use adaptive skin color modeling to overcome the varying skin color problem. Most methods for tracking skin regions in videos utilize the correlation between contiguous frames. This paper proposes a new approach for detecting skin in a single image. This approach uses a local skin model to shift a globally trained skin model to adapt the final skin model to the current image. Experimental results show that the proposed method can achieve better accuracy. Two improvements for speeding up the processing are also discussed.     

Coordination control of greenhouse environmental factors

Optimal control of greenhouse climate is one of the key techniques in digital agriculture. Greenhouse climate, a nonlinear and uncertain system, consists of several major environmental factors such as temperature, humidity, light intensity, and CO₂ concentration. Due to the complex coupled correlations, it is a challenge to achieve coordination control of greenhouse environmental factors. This paper proposes a model-free coordination control approach for greenhouse environmental factors based on Q-learning. Coordination control policy is found through systematic interaction with the dynamic environment to achieve optimal control for greenhouse climate with the control cost constraints. In order to decrease systematic trial-and-error risk and reduce the computational complexity in Q-learning algorithm, case-based reasoning (CBR) is seamlessly incorporated into the Q-learning process. The experimental results demonstrate that this approach is practical, highly effective and efficient.     

汉字识别方法综述

分析汉字结构特征及其演变,在汉字字库及汉字识别目标变化的基础上,结合汉字识别过程,研究与比较贝叶斯决策、二次判别函数法、隐马尔科夫模型、神经网络分类器以及支持向量机这5种汉字识别方法的优缺点及适用范围。结果表明,汉字识别方法仍需要在识别目标及特征提取、训练字符集准备方面进行改进。     

基于网格的混合神经网络计算平台研究与实现

为了模仿人脑的复杂功能,把各种相关类型的神经网络组织起来,形成一个大规模混合神经网络.根据此需求,使用自主研发的LabGrid技术开发了一个基于网格的混合神经网络计算平台,利用该平台设计了一种新的混合神经网络分类系统来对该平台进行测试.测试结果表明,该平台具有较高的效率和良好的容错性.与其它分类系统比较可知,该分类系统有较高的准确率,从而证明了模仿人脑建立大规模混合神经网络分类系统的可行性和有效性.     

基于模糊控制的驾驶疲劳检测

提出了一种判断疲劳程度的新方法,通过检测眼睛、嘴巴状态和头部位置等能够反应疲劳的生理特征,利用模糊控制器推理,得到人的疲劳状态的数值表示,改善了疲劳或者非疲劳的二值表示形式.通过肤色识别和阚值选取等方法得到人眼检测,进行边界提取的结果优于边缘检测.利用fisher线性分类器进行嘴唇和肤色的分类,提高了检测速度.模糊推理更精确的反应了人的疲劳程度,实验结果表明了检测方法的有效性和可信性.     

一种改进的AdaBoost算法在人脸检测系统中的应用

针对复杂背景条件下人脸检测的检测率低、速度慢的问题,提出了一种改进的AdaBoost算法,与遗传算法相结合,产生了一种识别率高、泛化能力好的强分类器,文中称之为GA-AdaBoost算法。该算法首先训练多个支持向量机作为弱分类器,然后用AdaBoost算法将多个弱分类器组合成一个强分类器,在组合的同时采用遗传算法对各弱分类器的权值进行全局寻优。最后,通过试验与传统AdaBoost进行对比,表明了该算法具有识别率高和速度快的优越性。     

一种基于案例属性特征的区间相似度改进算法研究

基于案例推理的关键是确定问题案例与案例库中的案例属性特征的相似度,针对相似度的算法问题,在分析比较多种已有算法的基础上,提出了一种解决感知信息和案例库中的案例属性特征均为不确定或者不完整信息的区间相似度计算模型,论证了该模型的计算结果适用于特定区间内微小区间相似的问题。结合某型艇抗沉辅助决策系统的预案推理问题,对已有的算法求解结果进行比较,证实了该模型提高了案例相似度计算结果灵敏度。