在樹莓派上建立一個最簡單手寫體識別系統(二)
阿新 • • 發佈:2018-12-31
首先得先把opencv安裝上。
在PC上我使用的是anaconda,直接輸入:
conda install --channel https://conda.anaconda.org/menpo opencv3
測試程式碼:
import cv2
print(cv2.__version__)
這一步真簡單,網上也到處能搜到,我這裡就是記個筆記。
第二步,使用opencv來讀取影象:
import cv2
from matplotlib import pyplot as plt
#讀取影象
img=cv2.imread('Test3.jpg',cv2.IMREAD_COLOR)
#顯示影象的方法1
cv2.imshow('Test3',img)
k=cv2.waitKey(0)
#顯示影象的方法2
#首先得把GRG調換個個頭
img3=img[:,:,::-1]
plt.imshow(img3)
plt.show()
為了適合Mnist模型,我們還得轉成灰度圖,反色,並縮放到28*28
img=cv2.imread('Test3.jpg',cv2.IMREAD_COLOR)
img =cv2.resize(img, (28, 28), interpolation=cv2.INTER_CUBIC)
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
img = 0xFF-img
cv2.imshow('Test3',img)
k=cv2.waitKey(0)
下面讀取w、b引數,還得把輸入影象畫素歸一化
然後直接計做矩陣乘加、softmax輸出:
import cv2
from matplotlib import pyplot as plt
img=cv2.imread('Test3.jpg',cv2.IMREAD_COLOR)
img =cv2.resize(img, (28, 28), interpolation=cv2.INTER_CUBIC)
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
img = 0xFF -img
img = img/255
img = img.reshape(784)
testx = [img]
from numpy import *;
import numpy as np; #這個方式使用numpy的函式時,需要以np.開頭。
import math
import pandas as pd
from pandas import Series,DataFrame
data = pd.read_csv('mnist_w.csv')
#print(data.shape)
usew = data.values
data = pd.read_csv('mnist_b.csv')
useb = data.values
useb = useb.reshape(1,10)#必須確定大小,否則會出錯
a1 = mat(img);
a2 = mat(usew);
a4 = a1*a2+useb
#計算softmax
etab=[0,0,0,0,0,0,0,0,0,0]
for i in range(0,10):
t = a4[0:,i]
t = t.tolist()[0]
t = t[0]
etab[i] = math.exp(t)
a = sum(etab)
for i in range(0,10):
t = etab[i]
t = t/a
print(t)
print ( "預測值=",etab.index(max(etab)) )
最後得到輸出:
0.0303718806336
0.0306637445764
0.0985644947741
0.496698573451
0.0684541058624
0.193191997366
0.0177791782552
0.0252268936033
0.0210481976761
0.0180009338012
預測值= 3
明顯,預測值為3,並且結果和直接呼叫tensorflow一致!