论文总字数：38854字

摘 要

 
  随着人工智能应用在各个领域的普及使用，智慧医疗也成为了人工智能应用的兵家必争之地。在眼科疾病方面，尤其是白内障识别方面，深度学习医学图像识别分析得到了发展和应用。如今全世界人口老龄化的日益加速，白内障作为一种在中老年人群中常见的眼科疾病，它的诊断治疗受到科学界人士的众多关注。一款提供准确高效的高性能白内障识别移动端系统，方便普通患者尤其是边缘地区患者低成本的诊断白内障是人工智能落地应用的终极目标。而方便快捷的图像识别智能手机APP可以帮助医生自动识别患者眼睛的患病情况，减轻医生压力，降低诊断失误概率。计算机的深度神经学习网络在白内障识别与分析、辅助医疗诊断方面有着非常重要的现实价值意义。在人工智能应用发展即将爆发的前夜，相信人工智能在手机端的应用会给智慧医疗行业注入新活力。目的 使用深度学习神经网络算法提取眼睛部分图像的白内障特征，生成一个对白内障图像自动分类的网络模型，应用于手机APP中对白内障图片进行智能识别分类。方法 1）在主机端训练完成对白内障的识别神经网络训练，RCNN图像目标检测识别框架使用ResNet-101作为网络模型从而实现对白内障的检测和识别。本文在神经网络训练完成后将生成的神经网络文件转化为TFLITE文件，实现在移动端对白内障的检测。2）实现安卓应用开发，利用TensorflowLite框架构建一个Android应用，在AndroidStudio软件中配置opencv的环境，使用opencv对接TensorflowLite框架，移植经过训练的白内障图像检测的卷积神经网络进行白内障图片识别，在安卓应用软件上实现对白内障的检测识别。结果 1）使用深度学习方法构件的分类器在白内障识别任务方面达到了可用级别。2）生成安卓应用软件可以使用的白内障检测APP。结论 基于深度学习的白内障识别算法有较好的识别效果，应用于移动端降低了检测成本，提高了检测效率。在未来具有很重要的实用意义，给医生和广大的白内障患者带来福音，该方法在智慧医疗领域具有潜在的应用前景。

关键词:人工智能；白内障；安卓应用；眼底图像；深度学习；

Abstract

With the widespread use of artificial intelligence applications in various fields, smart healthcare has also become a battlefield for artificial intelligence applications. In ophthalmic diseases, especially in cataract recognition, deep learning medical image recognition analysis has been developed and applied. Now that the aging of the world's population is accelerating, cataract is a common eye disease in the middle-aged and elderly population, and its diagnosis and treatment have received much attention from the scientific community. A mobile terminal system that provides accurate and efficient high-performance cataract recognition to facilitate the low-cost diagnosis of cataracts for ordinary patients, especially those in marginal areas, is the ultimate goal of artificial intelligence landing applications. The convenient and fast image recognition smartphone APP can help doctors automatically identify the patient's eye disease, reduce doctor pressure, and reduce the probability of diagnostic errors. The computer's deep neural learning network has very important practical value in cataract recognition and analysis, and auxiliary medical diagnosis. On the eve of the forthcoming development of artificial intelligence applications, I believe that the application of artificial intelligence on mobile phones will inject new vitality into the smart medical industry. Objective To use deep learning algorithms to automatically extract the cataract features of retinal fundus images. The component cataract automatic classifier is applied to the mobile app for intelligent recognition and classification of cataract pictures. Method 1) Complete the neural network training for road target detection and recognition on the computer client. The RCNN target detection framework uses the ResNet-101 framework as the backbone network for target detection and recognition. This paper uses an improved RCNN detection framework to convert the generated neural network file into a TFLITE file to realize the detection of cataracts on the mobile terminal. 2) To achieve Android application development, use the TensorflowLite framework to build an Android application, configure the opencv environment in the AndroidStudio software, use the opencv to connect to the TensorflowLite framework, and transplant the trained convolutional neural network for cataract image detection for cataract image recognition. The software realizes the detection and recognition of cataract. Results 1) Classifiers using deep learning methods have reached usable levels in cataract recognition tasks. 2) Generate a cataract detection app that can be used by Android application software. Conclusion The cataract recognition algorithm based on deep learning has a better recognition effect, and it is applied to the mobile terminal to reduce the detection cost and improve the detection efficiency. It has very important practical significance in the future and will bring gospel to doctors and the majority of cataract patients. This method has potential application prospects in the field of smart medicine.

Keywords: Artificial intelligence; cataract; Android APP; fundus image; deep learning;

目录

摘要 I

Abstract II

第一章 绪论 1

1.1选题背景 1

1.2研究目的 2

1.3研究意义 2

1.4研究内容 3

1.5研究方法 4

1.6文献综述 4

1.7本章总结 6

第2章 基于卷积神经网络的图像检测算法 6

2.1人工神经网络 7

2.2卷积神经网络 10

2.3基于卷积神经网络的目标检测算法 11

2.4 keras深度学习框架 16

2.5本章总结 17

第3章 基于R-CNN模型的白内障识别 18

3.4本章总结 27

第4章 安卓应用的开发 28

4.1前言 28

4.2安卓应用的开发 28

4.3系统设计与实现 32

4.4本章总结 38

第5章 总结和展望 39

5.1论文工作总结 39

5.2未来工作展望 39

参考文献 43

第一章 绪论

1.1选题背景

白内障是指眼睛晶状体变得混浊，从而导致视力下降。白内障正常情况下发展速度比较缓慢，并且在大多数情况下会影响患者的一只或两只眼睛。症状可能包括褪色，模糊或双重视力，光线周围的光晕，明亮的灯光带来的麻烦以及晚上看不清。这可能会严重影响患者正常生活。由白内障引起的视力低下也可能导致跌倒和抑郁的风险增加。全世界所有失明病例的一半都是白内障导致的并且33％的视力障碍起因也是白内障。

白内障最常见的原因是衰老，但也可能是由于外伤或辐射暴露而产生的，白内障是从出生时就可能会出现的，或者由于其他问题而在眼科手术后发生的。危险因素包括糖尿病，吸烟，长时间暴露在阳光下和酒精中。主要的致病机制涉及到蛋白质或黄棕色色素团块在晶状体中的积累，从而减少了光向眼后部视网膜的透射。通常的诊断是通过眼睛检查。

由于白内障这种疾病，给大约有3600万人造成失明。在中国白内障患者过亿，在发展中国家白内障造成的失明约占万分之四而这个数字在发达国家是万分之零点四。白内障随着年龄的增长变得越来越易发。在中国，80岁以上的人群中有68％患有白内障。

请支付后下载全文，论文总字数：38854字