计算机视觉 opencv
Utilizing computer vision and machine learning to automatically detect and predict Parkinson’s disease based on geometric drawings (i.e., spirals and sign waves).
利用计算机视觉和机器学习，可以根据几何图形(即螺旋形和符号波)自动检测和预测帕金森氏病。
绘制螺旋线和波以检测帕金森氏病 (Drawing spirals and waves to detect Parkinson’s disease)
A 2017 study by Zham et al. found that it was possible to detect Parkinson’s by asking the patient to draw a spiral and then track:
Zham等人的2017年研究。 发现可以通过要求患者画出螺旋线然后进行跟踪来检测帕金森氏症：
- Speed of drawing 绘图速度
- Pen pressure 笔压
The researchers found that the drawing speed was slower and the pen pressure lower among Parkinson’s patients — this was especially pronounced for patients with a more acute/advanced forms of the disease.
研究人员发现， 帕金森氏症患者的绘画速度较慢，笔压较低 -对于这种疾病的急性/晚期患者尤其如此 。
We’ll be leveraging the fact that two of the most common Parkinson’s symptoms include tremors and muscle rigidity which directly impact the visual appearance of a hand drawn spiral and wave.
我们将利用以下事实：帕金森氏症最常见的两种症状包括震颤和肌肉僵硬，这直接影响到手绘螺旋和波浪形的视觉外观。
The variation in visual appearance will enable us to train a computer vision and machine learning algorithm to automatically detect Parkinson’s disease.
视觉外观的变化将使我们能够训练计算机视觉和机器学习算法，以自动检测帕金森氏病。
数据集 (Dataset)
The dataset itself consists of 204 images and is pre-split into a training set and a testing set, consisting of:
数据集本身包含204张图像，并被预先分为训练集和测试集，其中包括：
Spiral: 102 images, 72 training, and 30 testing
螺旋： 102张图像，72次训练和30次测试
Wave: 102 images, 72 training, and 30 testing
Wave： 102张图像，72次训练和30次测试
Here we would be applying two methods :
在这里，我们将应用两种方法：
- Apply computer vision and OpenCV to detect Parkinson’s based on geometric drawings and then train a machine learning model to classify them 应用计算机视觉和OpenCV根据几何图形检测帕金森氏病，然后训练机器学习模型对其进行分类
- Apply deep learning(Fastai) 应用深度学习(法泰)
需要包装 (Packages Required)
- OpenCV OpenCV
- NumPy NumPy
- Scikit-learn Scikit学习
- Scikit-image Scikit图片
- imutils 不实用
导入库 (Import Libraries)
定义函数以HOG(定向梯度直方图)方法量化波/螺旋图像 (Define a function to quantify a wave/spiral image with the HOG(Histogram of Oriented Gradients) method)
定义一个函数以接受数据集路径并返回所有要素数据和关联的类标签 (Define a function to accept a dataset path and returning all feature data and associated class labels)
提取训练和测试特征以获取螺旋图像 (Extract Train and Test Features for spiral images)
将标签编码为整数 (Encode the Labels into Integer)
使用RandomForest分类器训练模型并评估测试图像 (Train the model using RandomForest Classifier and evaluate on the test images)
遍历为每个线索生成的指标 (Loop over the metrics generated for each trail)
Output :
输出：
Validation Accuracy for Spiral images is 82%
螺旋图像的验证精度为82％
测试在随机图像上训练的模型： (Test the model trained on random Images:)
- randomly sample images from our testing set. 从我们的测试集中随机采样图像。
- Loop over the random image indices 循环遍历随机图像索引
- Image list will hold each wave or spiral image along with annotations added via OpenCV drawing functions. 图像列表将保存每个波形或螺旋图像，以及通过OpenCV绘图功能添加的注释。
for each image in the list automatically classify the image using our new HOG + Random Forest based classifier and add color-coded annotations
对于列表中的每个图像，使用我们新的基于HOG + Random Forest的分类器对图像进行自动分类，并添加颜色编码的注释
- Finally add the image list to opencv montage for visualization 最后将图像列表添加到opencv montage以进行可视化
The opencv build_montages function requires three arguments:
opencv build_montages函数需要三个参数：
image_list : This parameter is a list of images loaded via OpenCV.
image_list ： 此参数是通过OpenCV加载的图像的列表。
image_shape: A tuple containing the width and height of each image in the montage.
image_shape ： 一个元组，包含蒙太奇中每个图像的宽度和高度。
montage_shape: A second tuple, this one specifying the number of columns and rows in the montage. Here we indicate that our montage will have 5columns (5 images wide) and 5rows (5 images tall).
montage_shape ： 第二个元组，该元组指定蒙太奇中的列和行数。 在这里，我们指示蒙太奇将有5列(宽5幅图像)和5行(高5幅图像)。
The build_montages method returns a list of montage images in NumPy array format.
build_montages方法以NumPy数组格式返回蒙太奇图像的列表。
Note: Empty space in the montage will be filled with black pixels.
注意： 蒙太奇中的空白将被黑色像素填充。
The class label is colored green for “healthy” and red for “parkinsons”
班级标签的绿色表示 “健康” ，红色表示 “帕金森”
类似地训练波形图像模型并在测试图像上进行评估 (Similarly training the model for wave images and evaluating on the test Images)
Output :
输出：
Validation Accuracy for wave images is 65%
波浪图像的验证精度为65％
帕金森病分类器：使用Fastai (Parkinsons Disease Classifier : using Fastai)
导入所需的Fastai库 (Import required Fastai libraries)
设置图像路径 (Set image path)
启用变换(图像增强) (Enabling Transforms (Image Augmentation))
Transforms are passed on when creating the “ImageDataBunch” objects.Genearlly, you may enable the “Default” transforms by calling
创建“ ImageDataBunch”对象时会传递转换。通常，您可以通过调用启用“默认”转换
tfms = get_transforms()
tfms = get_transforms()
The tuple contating transforms has 2 lists nested.
元组连续转换有2个嵌套列表。
- One is for the training dataset. 一种是用于训练数据集。
- Second one is for the validation dataset that involves minimal transforms/just resizing. 第二个是用于验证数据集，该数据集涉及最少的转换/仅调整大小。
tfms [0] (tfms[0])
[RandTransform(tfm=TfmCrop (crop_pad), kwargs={‘row_pct’: (0, 1), ‘col_pct’: (0, 1), ‘padding_mode’: ‘reflection’}, p=1.0, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmAffine (dihedral_affine), kwargs={}, p=1.0, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmCoord (symmetric_warp), kwargs={‘magnitude’: (-0.2, 0.2)}, p=0.75, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmAffine (rotate), kwargs={‘degrees’: (-180, 180)}, p=0.75, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmAffine (zoom), kwargs={‘scale’: (1.0, 1.1), ‘row_pct’: (0, 1), ‘col_pct’: (0, 1)}, p=0.75, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmLighting (brightness), kwargs={‘change’: (0.4, 0.6)}, p=0.75, resolved={}, do_run=True, is_random=True, use_on_y=True), RandTransform(tfm=TfmLighting (contrast), kwargs={‘scale’: (0.8, 1.25)}, p=0.75, resolved={}, do_run=True, is_random=True, use_on_y=True)]
[RandTransform(tfm = TfmCrop (crop_pad)，kwargs = {'row_pct'：(0，1)，'col_pct'：(0，1)，'padding_mode'：'reflection'}，p = 1.0，已解决= {} ，do_run = True，is_random = True，use_on_y = True)，RandTransform(tfm = TfmAffine ( dihedral_affine) ，kwargs = {}，p = 1.0，resolve = {}，do_run = True，is_random = True，use_on_y = True)， RandTransform(tfm = TfmCoord (symmetric_warp)，kwargs = {'magnitude'：(- 0.2，0.2 )}，p = 0.75，resolve = {}，do_run = True，is_random = True，use_on_y = True)，RandTransform(tfm = TfmAffine (旋转 )，kwargs = {'度数：(-180，180)}，p = 0.75，已解析= {}，do_run = True，is_random = True，use_on_y = True)，RandTransform(tfm = TfmAffine(缩放 )， kwargs = {'scale'：(1.0，1.1)，'row_pct'：(0，1)，'col_pct'：(0，1)}，p = 0.75，resolve = {}，do_run = True，is_random = True ，use_on_y = True)，RandTransform(tfm = TfmLighting(亮度) ，kwargs = {'change'：( 0.4，0.6 )}，p = 0.75，resolve = {}，do_run = True，is_random = True，use_on_y = True) ，RandTransform(tfm = TfmLighting(contrast) ，kwargs = {'scale'：( 0.8，1.25 )}，p = 0.75，resolve = {}，do_run = True，is_random = True，use_on_y = True)]]
tfms [1] (tfms[1])
[RandTransform(tfm=TfmCrop (crop_pad), kwargs={}, p=1.0, resolved={}, do_run=True, is_random=True, use_on_y=True)]
[RandTransform(tfm = TfmCrop(crop_pad) ，kwargs = {}，p = 1.0，已解决= {}，do_run = True，is_random = True，use_on_y = True)]
使用fastai的ImageDataBunch类加载和准备数据 (Loading and preparing the data with fastai’s ImageDataBunch class)
类标签： (Class Labels:)
data.classes :Class Labels for the training samples : [‘healthy’, ‘parkinson’]
data.classes ： 训练样本的类别标签：['healthy'，'parkinson']
可视化数据 (Visualize the Data)
训练模型 (Train the Model)
找到最佳学习率 (Find the Optimum learning rate)
应用最佳学习率 (Applying optimum learning rate)
情节损失 (Plot Losses)
混淆度量以评估模型性能 (Confusion Metrics to evaluate the model performance)
保存第一个训练后的模型并找到训练和验证的准确性 (Save the first trained model and find the training and validation accuracy)
Here we could see that we could easily get a validation accuracy 0f 86% for Spiral images with only few lines of code as compared to the opencv+ ML code.
在这里我们可以看到，与opencv + ML代码相比，只需几行代码，就可以轻松获得螺旋图像的验证精度0f 86％。
可视化最大损失 (Visualize top losses)
虚拟化螺旋图像的预测 (Viualize Predictions for Spiral images)
波浪图像的训练模型 (Train Model for Wave Images)
显示结果 (Show Results)
确定最佳学习率 (Optimum Learning rate determined)
应用最佳学习率 (Applying Optimum Learning Rate)
评估训练波模型的性能 (Evaluate the performance of trained wave model)
训练和验证准确性 (Training and Validation Accuracies)
显示经过训练的模型的结果 (Show Results for the trained Model)
This validation accuracy is 83% for wave images with only few lines of code as compared to the opencv+ ML code.
与opencv + ML代码相比，仅几行代码的波形图像的验证精度为83％。
connect
连接
Reference:
参考：
Adrian Rosebrock的 开放式简历和深度学习 (Open CV and Deep Learning by Adrian Rosebrock)
翻译自: https://medium.com/@nayakpplaban/detecting-parkinsons-with-opencv-computer-vision-and-the-spiral-wave-test-e8de3b30f5e6
计算机视觉 opencv