[day-20] K-means 分類法分類影像

11th鐵人賽

disapear1997

2019-09-25 23:55:10

1173 瀏覽

分享至

前言

上一篇介紹了K-means的運作機制，接下來我們進入實作的部分。

分類流程

首先，因為我們是GIS實作，所以分類的對象是影像，因此我們希望可以做到以下的流程：

Load 照片為三維矩陣
Reshape為2維度矩陣
匯入K-means計算類別
將成果還原為三維矩陣
儲存並呈現

實作

我們使用的函式庫scikit-learn函式庫，有很多機器學習的工具，從上述這張地圖可以知道scikit-learn機器學習的使用時機。

import skimage
import skimage.io
import skimage.external
from sklearn import cluster
import numpy

def image_kmeans_classification(image_array,bands):
    image_array = numpy.asarray(image_array)
    row_number = image_array.shape[0]
    col_number = image_array.shape[1]
    
    sample = image_array.reshape((row_number*col_number,bands))
    
    kmeans_classifer = cluster.KMeans(n_clusters = 4,random_state=0).fit(sample)
    kmeans_result = kmeans_classifer.predict(sample)
    kmeans_result = (kmeans_result-numpy.min(kmeans_result))/(numpy.max(kmeans_result)-numpy.min(kmeans_result))*255
    image_array_classification = kmeans_result.reshape((row_number,col_number))
    return image_array_classification
    
def main():
    image_path = r"D:\Nantou County-Ren ai Township-D043-UAV ladir data\DEM TO TXT\37ne2_feature.tif"
    output_image_path = r"D:\Nantou County-Ren ai Township-D043-UAV ladir data\DEM TO TXT\37ne2_feature_result.tif"
    
    bands = 4
    
    
    image_array = skimage.io.imread(image_path)
    image_array_classification = image_kmeans_classification(image_array,bands)
    #(jpg﹑png) skimage.io.imsave(output_image_path,image_array_classification)
    skimage.external.tifffile.imsave(output_image_path,image_array_classification)
    
if __name__ == "__main__":
    main()