叶片形状分类

import skimage,math,numpy
from sklearn.decomposition import PCA
from skimage import io,transform,measure
from skimage.color import rgb2gray
from skimage.filters import threshold_otsu, threshold_sauvola

path='F:\数据挖掘原理\dataset\题目三\圆叶\*.jpg'
coll1 = io.ImageCollection(path)
path='F:\数据挖掘原理\dataset\题目三\尖叶\*.jpg'
coll2 = io.ImageCollection(path)

feature1=[]
#提取圆叶的特征值
for num in range(0,len(coll1)):
    #转换为二值图
    origin_img=coll1[num]
    origin_img = transform.rescale(origin_img, 0.1)
    gray_img=rgb2gray(origin_img)
    otsu_threshold = threshold_otsu(gray_img)
    gray_img = skimage.img_as_ubyte(gray_img)
    bin_image = gray_img > otsu_threshold
    
    labels=measure.label(bin_image)#连通区域标记
    
    repr=measure.regionprops(labels)#获取连通域的特征属性
    
    max_area=0
    yezi=repr[0]
    for i in range(0,len(repr)):
        if(repr[i].area>=max_area):
            max_area=repr[i].area
            yezi=repr[i]
    Cir_ex=(4*math.pi*max_area)/(yezi.perimeter*yezi.perimeter)
    Rec_ex=yezi.extent
    #得到特征圆形度和矩形度
    
    a=numpy.zeros([bin_image.shape[1]])
    bin_image1=numpy.row_stack((a,bin_image,a))
    a1=numpy.zeros([bin_image1.shape[0]])
    a1=a1.T
    bin_image1=numpy.c_[a1,bin_image1,a1]
    max_height=0
    outline=[]
    for y in range(1,bin_image1.shape[0]-2):
        for x in range(1,bin_image1.shape[1]-1):
            if((bin_image1[y][x-1]==0 and bin_image1[y][x]==1)or(bin_image1[y][x]==1 and bin_image1[y][x+1]==0) or( bin_image1[y-1][x]==0 and bin_image1[y][x]==1 )or(bin_image1[y+1][x]==0 and bin_image1[y][x]==1)):
                outline.append([x,y])                                                     # 提取边缘坐标
                
    for i in range(0,len(outline)):
        for j in range(i+1,len(outline)):
            length=math.sqrt((outline[j][0]-outline[i][0])*(outline[j][0]-outline[i][0])+(outline[j][1]-outline[i][1])*(outline[j][1]-outline[i][1]))
            if(length>max_height):
                max_height=length                                                        # 提取叶子的最大长度
                x1=outline[i][0]
                x2=outline[j][0]
                y1=outline[i][1]
                y2=outline[j][1]
    max_width=0
    min=math.fabs((outline[1][0]-outline[0][0])*(x2-x1)+(outline[1][1]-outline[0][1])*(y2-y1))
    for i in range(0,len(outline)-1):
        for j in range(i+1,len(outline)):
            if(math.fabs((outline[j][0]-outline[i][0])*(x2-x1)+(outline[j][1]-outline[i][1])*(y2-y1))max_width):
                    max_width=width                                                      # 提取叶子的最大宽度 
                    x3=outline[i][0]
                    x4=outline[j][0]
                    y3=outline[i][1]
                    y4=outline[j][1]
                    
    wh_cp=max_width/max_height
    feature1.append([Cir_ex,Rec_ex,wh_cp])

feature2=[]
#提取尖叶的特征值
for num in range(0,len(coll2)):
    #转换为二值图
    origin_img=coll2[num]
    origin_img = transform.rescale(origin_img, 0.1)
    gray_img=rgb2gray(origin_img)
    otsu_threshold = threshold_otsu(gray_img)
    gray_img = skimage.img_as_ubyte(gray_img)
    bin_image = gray_img > otsu_threshold
    
    labels=measure.label(bin_image)#连通区域标记
    
    repr=measure.regionprops(labels)#获取连通域的特征属性
    
    max_area=0
    yezi=repr[0]
    for i in range(0,len(repr)):
        if(repr[i].area>=max_area):
            max_area=repr[i].area
            yezi=repr[i]
    Cir_ex=(4*math.pi*max_area)/(yezi.perimeter*yezi.perimeter)
    Rec_ex=yezi.extent
    #得到特征圆形度和矩形度
    
    a=numpy.zeros([bin_image.shape[1]])
    bin_image1=numpy.row_stack((a,bin_image,a))
    a1=numpy.zeros([bin_image1.shape[0]])
    a1=a1.T
    bin_image1=numpy.c_[a1,bin_image1,a1]
    max_height=0
    outline=[]
    for y in range(1,bin_image1.shape[0]-2):
        for x in range(1,bin_image1.shape[1]-1):
            if((bin_image1[y][x-1]==0 and bin_image1[y][x]==1)or(bin_image1[y][x]==1 and bin_image1[y][x+1]==0) or( bin_image1[y-1][x]==0 and bin_image1[y][x]==1 )or(bin_image1[y+1][x]==0 and bin_image1[y][x]==1)):
                outline.append([x,y])                                                     # 提取边缘坐标
                
    for i in range(0,len(outline)):
        for j in range(i+1,len(outline)):
            length=math.sqrt((outline[j][0]-outline[i][0])*(outline[j][0]-outline[i][0])+(outline[j][1]-outline[i][1])*(outline[j][1]-outline[i][1]))
            if(length>max_height):
                max_height=length                                                        # 提取叶子的最大长度
                x1=outline[i][0]
                x2=outline[j][0]
                y1=outline[i][1]
                y2=outline[j][1]
    max_width=0
    min=math.fabs((outline[1][0]-outline[0][0])*(x2-x1)+(outline[1][1]-outline[0][1])*(y2-y1))
    for i in range(0,len(outline)-1):
        for j in range(i+1,len(outline)):
            if(math.fabs((outline[j][0]-outline[i][0])*(x2-x1)+(outline[j][1]-outline[i][1])*(y2-y1))max_width):
                    max_width=width                                                      # 提取叶子的最大宽度 
                    x3=outline[i][0]
                    x4=outline[j][0]
                    y3=outline[i][1]
                    y4=outline[j][1]
                    
    wh_cp=max_width/max_height
    feature2.append([Cir_ex,Rec_ex,wh_cp])

#特征主成分分析
feature=feature1+feature2
pca=PCA(n_components=3)
pca.fit(feature)
weight=pca.explained_variance_ratio_
print(pca.explained_variance_ratio_)

#得到圆叶判别公式
#Fx=特征1*权重+特征2*权重+特征3*权重
Fx1=[]
for fe in feature1:
    Fx=fe[0]*weight[0]+fe[1]*weight[1]+fe[2]*weight[2]
    Fx1.append(Fx)
m1=numpy.max(Fx1)
m2=numpy.min(Fx1)
print(Fx1)
print(m1)
print(m2)

#尖叶叶特征主成分分析
pca=PCA(n_components=3)
pca.fit(feature2)
#得到尖叶判别公式
#Fx=特征1*权重+特征2*权重+特征3*权重
Fx2=[]
for fe in feature2:
    Fx=fe[0]*weight[0]+fe[1]*weight[1]+fe[2]*weight[2]
    Fx2.append(Fx)
print(Fx2)
l1=numpy.max(Fx2)
l2=numpy.min(Fx2)
print(l1)
print(l2)

path='F:\数据挖掘原理\dataset\题目三\叶子\*.jpg'
coll3 = io.ImageCollection(path)
#提取叶子的特征值
i=1
j=1
for num in range(0,len(coll3)):
    #转换为二值图
    origin_img=coll3[num]
    origin_img = transform.rescale(origin_img, 0.1)
    gray_img=rgb2gray(origin_img)
    otsu_threshold = threshold_otsu(gray_img)
    gray_img = skimage.img_as_ubyte(gray_img)
    bin_image = gray_img > otsu_threshold
    
    labels=measure.label(bin_image)#连通区域标记
    
    repr=measure.regionprops(labels)#获取连通域的特征属性
    
    max_area=0
    yezi=repr[0]
    for i in range(0,len(repr)):
        if(repr[i].area>=max_area):
            max_area=repr[i].area
            yezi=repr[i]
    Cir_ex=(4*math.pi*max_area)/(yezi.perimeter*yezi.perimeter)
    Rec_ex=yezi.extent
    #得到特征圆形度和矩形度
    
    a=numpy.zeros([bin_image.shape[1]])
    bin_image1=numpy.row_stack((a,bin_image,a))
    a1=numpy.zeros([bin_image1.shape[0]])
    a1=a1.T
    bin_image1=numpy.c_[a1,bin_image1,a1]
    max_height=0
    outline=[]
    for y in range(1,bin_image1.shape[0]-2):
        for x in range(1,bin_image1.shape[1]-1):
            if((bin_image1[y][x-1]==0 and bin_image1[y][x]==1)or(bin_image1[y][x]==1 and bin_image1[y][x+1]==0) or( bin_image1[y-1][x]==0 and bin_image1[y][x]==1 )or(bin_image1[y+1][x]==0 and bin_image1[y][x]==1)):
                outline.append([x,y])                                                     # 提取边缘坐标
                
    for i in range(0,len(outline)):
        for j in range(i+1,len(outline)):
            length=math.sqrt((outline[j][0]-outline[i][0])*(outline[j][0]-outline[i][0])+(outline[j][1]-outline[i][1])*(outline[j][1]-outline[i][1]))
            if(length>max_height):
                max_height=length                                                        # 提取叶子的最大长度
                x1=outline[i][0]
                x2=outline[j][0]
                y1=outline[i][1]
                y2=outline[j][1]
    max_width=0
    min_=math.fabs((outline[1][0]-outline[0][0])*(x2-x1)+(outline[1][1]-outline[0][1])*(y2-y1))
    for i in range(0,len(outline)-1):
        for j in range(i+1,len(outline)):
            if(math.fabs((outline[j][0]-outline[i][0])*(x2-x1)+(outline[j][1]-outline[i][1])*(y2-y1))max_width):
                    max_width=width                                                      # 提取叶子的最大宽度 
                    x3=outline[i][0]
                    x4=outline[j][0]
                    y3=outline[i][1]
                    y4=outline[j][1]
                    
    wh_cp=max_width/max_height
    if((Cir_ex*weight[0]+Rec_ex*weight[1]+wh_cp*weight[2])>=m2 and (Cir_ex*weight[0]+Rec_ex*weight[1]+wh_cp*weight[2])<=m1):
        print("圆叶",end='')
        str1="yuan"+str(i)+".jpg"
        io.imsave("F:\数据挖掘原理\image_classification_demo-master\叶子\圆叶\{0}".format(str1),coll3[num])
        i=i+1
    elif((Cir_ex*weight[0]+Rec_ex*weight[1]+wh_cp*weight[2])>=l2 and (Cir_ex*weight[0]+Rec_ex*weight[1]+wh_cp*weight[2])<=l1):
        print("尖叶",end=' ')
        str2="jian"+str(j)+".jpg"
        io.imsave("F:\数据挖掘原理\image_classification_demo-master\叶子\尖叶\{0}".format(str2),coll3[num])
        j=j+1
    else:
        print("无法识别")