dnn图片分类

dnn_image/convert_img.py

@@ -0,0 +1,42 @@
+# -*- encoding:utf-8 -*-
+import numpy as np
+from PIL import Image
+import random
+
+
+def read_data(path):
+	with open(path) as f :
+		lines=f.readlines()
+	lines=random.sample(lines,int(len(lines)/1000))
+	lines=[eval(line.strip()) for line in lines]
+	X,Y=zip(*lines)
+	X=np.array(X)
+	X=X.reshape(-1,28*28)
+	Y=np.array(Y)
+	return X,Y
+
+
+def plot(x,width,height,path):
+	img=[[0 for _ in range(0,width) ] for _ in range(0,height)]
+	for i in range(0,height):
+		for j in range(0,width):
+			img[i][j]=x[i*height+j]
+	img=np.array(img).astype('uint8')
+	new_im = Image.fromarray(img)
+	new_im.save(path)
+
+
+train_x,train_y = read_data("train_data")
+train_x_10,train_y_10 = read_data("train_data_10")
+train_x = np.concatenate((train_x,train_x_10))
+train_y = np.concatenate((train_y,train_y_10))
+count=0
+
+
+for i in range(0,len(train_x)):
+	print(count)
+	path="img/{}-{}.png".format(train_y[i],count)
+	print(path)
+	count+=1
+	plot(train_x[i],28,28,path)
+

dnn_image/generate_random_data.py

@@ -0,0 +1,15 @@
+# -*- encoding:utf-8 -*-
+import numpy as np
+import random
+def generate():
+	X=[[random.randint(0,255) for i in range(0,28)] for j in range(0,28)]
+	X=[[0 if s <128 else s for s in ss] for ss in X]
+	return str([X,10])
+train_num=6000
+test_num=1000
+with open("train_data_10","w") as f :
+	train_data=[ generate() for _ in range(0,train_num)]
+	f.writelines("\n".join(train_data))
+with open("test_data_10","w") as f :
+	train_data=[ generate() for _ in range(0,test_num)]
+	f.writelines("\n".join(train_data))

dnn_image/get_data.py

@@ -0,0 +1,26 @@
+# -*- encoding:utf-8 -*-
+import numpy as np
+
+
+def load_data(path='mnist.npz'):
+    f = np.load(path)
+    x_train, y_train = f['x_train'], f['y_train']
+    x_test, y_test = f['x_test'], f['y_test']
+    print(len(y_train))
+    print(len(y_test))
+    f.close()
+    return (x_train, y_train), (x_test, y_test)
+
+
+print("开始读取数据")
+(train_X, train_y), (test_X, test_y) = load_data()
+print("读取结束")
+
+train_data=zip(train_X,train_y)
+train_data=[str([x.tolist(),y]) for [x,y] in train_data]
+test_data=zip(test_X,test_y)
+test_data=[str([x.tolist(),y]) for [x,y] in test_data]
+with open("train_data","w") as f:
+	f.writelines("\n".join(train_data))
+with open("test_data","w") as f:
+	f.writelines("\n".join(test_data))

dnn_image/predict.py

@@ -0,0 +1,30 @@
+from keras.models import load_model
+import numpy as np
+from keras.utils import np_utils
+
+
+def read_data(path):
+	with open(path) as f :
+		lines=f.readlines()[0:3]
+	lines=[eval(line.strip()) for line in lines]
+	X,Y=zip(*lines)
+	X=np.array(X)
+	X=1.0*X/256
+	X=X.reshape(-1,28*28)
+	Y=np.array(Y)
+	Y=np_utils.to_categorical(Y,num_classes)
+	return X,Y
+
+
+num_classes=11
+X,Y=read_data("test_data")
+model = load_model('model')  
+results=model.predict(X)
+
+for result,y in zip(results,Y):
+	print(result)
+	print(y)
+
+
+
+

dnn_image/train_dnn_seq.py

@@ -0,0 +1,47 @@
+# -*- encoding:utf-8 -*-
+import numpy
+from tensorflow.keras.optimizers import SGD, Adam
+from keras.models import Sequential
+from keras.layers import Dense
+from keras.layers import Dropout
+from keras.layers import Flatten
+from keras.utils import np_utils
+from keras import backend as K
+from keras.utils import np_utils
+import numpy as np
+import random
+
+
+
+def read_data(path):
+	with open(path) as f :
+		lines=f.readlines()
+	lines=random.sample(lines,int(len(lines)/10))
+	lines=[eval(line.strip()) for line in lines]
+	X,Y=zip(*lines)
+	X=np.array(X)
+	X=1.0*X/256
+	X=X.reshape(-1,28*28)#784
+	Y=np.array(Y)
+	Y=np_utils.to_categorical(Y,num_classes)
+	return X,Y
+		
+num_classes=11
+train_x,train_y=read_data("train_data")
+train_x_10,train_y_10=read_data("train_data_10")#reject data
+train_x=np.concatenate((train_x,train_x_10))
+train_y=np.concatenate((train_y,train_y_10))
+model = Sequential()
+#input_shape=(输入长，输入宽，输入通道数)
+model.add(Dense(units=256,input_dim=28*28, activation='relu',use_bias=True))
+model.add(Dense(units=128,input_dim=256,activation='relu',use_bias=True))
+model.add(Dense(units=128,input_dim=256,activation='relu',use_bias=True))
+model.add(Dense(units=num_classes,input_dim=128,use_bias=True, activation='softmax'))
+# Compile model
+model.compile(loss='categorical_crossentropy', optimizer="adam", metrics=['accuracy'])
+
+model.fit(train_x, train_y, batch_size=50, epochs=10,shuffle=True)
+model.save("model")
+
+
+