多光谱遥感分类（三）:CNN提取特征+RF分类

继上篇。

CNN特征+其他分类器

提取特征。

import gdal
from keras import Model
from keras.models import load_model
import numpy as np
import cv2
import os
import pandas as pd


model=load_model("output/classify3_vgg16_pro.h5")
# model.trainable=False
dense1_layer_model = Model(inputs=model.input,outputs=model.layers[-2].output)
dense1_layer_model.trainable=False
dense1_layer_model.summary()

dataset = gdal.Open(r"D:/test.tif")
size=128
def get_cell( pos_x, pos_y):
    try:
        output = []
        for i in [3,2,1]:
            band = dataset.GetRasterBand(i)
            t = band.ReadAsArray(int(pos_x - size / 2), int(pos_y - size / 2), size, size)
            output.append(t)
        img2 = np.moveaxis(np.array(output, dtype=np.uint8), 0, 2)
        # print(img2.shape)
        # self.showImg(img2)
    except:
        return None
    return img2


def get_cells():
    res=[]

    fea = pd.read_csv("tr_1.txt", header=None)  # 读取坐标-特征文件
    fea_len = len(fea)
    print("fea length: %d" % fea_len)

    for i in range(fea_len):
        temp = fea.iloc[i, :].values
        img = get_cell(temp[1], temp[0])
        if img is None:
            continue
        input=[]
        input.append(img)
        input=np.array(input)
        pred_y=dense1_layer_model.predict(input).reshape(-1)
        # 并上其他特征
        res.append(np.append(pred_y,temp[2:]))
        if(i%1000==0):
            print("%d fininsh." % i)
    res=np.array(res)
    df=pd.DataFrame(res)
    print(df.shape)
    df.to_csv("output/1024&108feather.txt",header=False,index=None)

get_cells()

使用其他分类器分类（RF）

import  pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder,StandardScaler
from sklearn.ensemble import RandomForestClassifier

data=pd.read_csv("output/1024&108feather.txt",header=None)
print(data.iloc[:5])

X=data.iloc[:,:-2]
Y=data.iloc[:,-2]
scaler = StandardScaler().fit(X)
X_scaler=scaler.transform(X)
label_t=LabelEncoder().fit(Y)
Y_t=label_t.transform(Y)

rf = RandomForestClassifier(n_estimators=100, max_depth=20)
rf.fit(X_scaler,Y_t)

data_test=pd.read_csv('output/test_1024&108feather',header=None)
X_test=data_test.iloc[:,:-2]
Y_test=data_test.iloc[:,-2]
X_test_scaler=scaler.transform(X_test)
Y_test_t=label_t.transform(Y_test)

score=rf.score(X_test_scaler,Y_test_t)
print(score)

作为对比，直接使用其他特征使用其他分类器分类。

import  pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder,StandardScaler
from sklearn.ensemble import RandomForestClassifier

data=pd.read_csv(r"D:/train1.txt",header=None, sep='\t')
print(data.iloc[:5])

X=data.iloc[:,:-2]
Y=data.iloc[:,-2]
scaler = StandardScaler().fit(X)
X_scaler=scaler.transform(X)
label_t=LabelEncoder().fit(Y)
Y_t=label_t.transform(Y)

rf = RandomForestClassifier(n_estimators=100, max_depth=20)
rf.fit(X_scaler,Y_t)

data_test=pd.read_csv(r"D:/test1.txt",header=None, sep='\t')
X_test=data_test.iloc[:,:-2]
Y_test=data_test.iloc[:,-2]
X_test_scaler=scaler.transform(X_test)
Y_test_t=label_t.transform(Y_test)

score=rf.score(X_test_scaler,Y_test_t)
print(score)