def test_image_2d():
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
# 图像数据（灰度）
file_name = "../picture/base_test/dog1.bmp"
show_file(file_name)
gray, err = gray_file(file_name)
show_data(gray, ImageDataType.GRAY)
# 卷积
con = Convolution()
# con = MeanPooling()
# 输入信息 x
x = gray
# 滤波器 w（平滑去噪）
w = np.asarray([[1 / 16, 1 / 8, 1 / 16], [1 / 8, 1 / 4, 1 / 8],
[1 / 16, 1 / 8, 1 / 16]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.GRAY)
# 滤波器 w（提取边缘特征）
w = np.asarray([[0, 1, 0], [1, -4, 1], [0, 1, 0]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.GRAY)
# 滤波器 w（提取边缘特征）
w = np.asarray([[0, 1, 1], [-1, 0, 1], [-1, -1, 0]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.GRAY)
def test_image_3d():
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
# 图像数据
file_name = "../picture/base_test/dog1.bmp"
show_file(file_name)
data, image_data_type, err = get_data(file_name)
# 卷积
con = Convolution()
# con = MaxPooling()
# 输入信息 x
x = data
# 滤波器 w（平滑去噪）
w = np.asarray([[[1 / 16, 1 / 16, 1 / 16], [1 / 8, 1 / 8, 1 / 8],
[1 / 16, 1 / 16, 1 / 16]],
[[1 / 8, 1 / 8, 1 / 8], [1 / 4, 1 / 4, 1 / 4],
[1 / 8, 1 / 8, 1 / 8]],
[[1 / 16, 1 / 16, 1 / 16], [1 / 8, 1 / 8, 1 / 8],
[1 / 16, 1 / 16, 1 / 16]]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.RGB)
# 滤波器 w（提取边缘特征）
w = np.asarray([[[0, 0, 0], [1, 1, 1], [0, 0, 0]],
[[1, 1, 1], [-4, -4, -4], [1, 1, 1]],
[[0, 0, 0], [1, 1, 1], [0, 0, 0]]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.RGB)
# 滤波器 w（提取边缘特征）
w = np.asarray([[[0, 0, 0], [1, 1, 1], [1, 1, 1]],
[[-1, -1, -1], [0, 0, 0], [1, 1, 1]],
[[-1, -1, -1], [-1, -1, -1], [0, 0, 0]]])
y, err = con.convolution(w, x, Reversal.REV, ConvolutionType.Other, 3, 0,
0)
show_data(y, ImageDataType.GRAY)
def test_cvl_nn_without_train():
# 1. 构建训练输入样本
# 图像文件名
file_name = "../picture/number/1_sx.bmp"
# 取灰度值
data, err = gray_file(file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
gray = my_image.array_3_2(data)
show_data(gray, ImageDataType.GRAY)
# 归一化
sx = my_image.normalize(data, my_image.NormalizationType.NORMAL)
# 显示归一化灰度图像
gray = my_image.array_3_2(sx)
show_data(gray, ImageDataType.GRAY)
# 训练输入样本
sx_list = list()
sx_list.append(sx)
# 2. 构建训练输出样本
# 图像数据
file_name = "../picture/number/8_sy_2.bmp"
# 取灰度值
data, err = gray_file(file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
gray = my_image.array_3_2(data)
# show_data(gray, ImageDataType.GRAY)
# 归一化
sy = my_image.normalize(data, my_image.NormalizationType.NORMAL)
# 显示归一化灰度图像
gray = my_image.array_3_2(sy)
show_data(gray, ImageDataType.GRAY)
# 训练输出样本
sy_list = list()
sy_list.append(sy)
# 3. 卷积神经网络的基本参数
''' W, B 参数 begin '''
# 每一层 w、B 参数，w 是个 matrix，b 是个 vector（数据类型也是一个 matrix）
W = list()
B = list()
# 第1层
w0 = np.asarray([[[50.113265], [93.826943], [115.373675]],
[[59.189640], [119.609671], [104.736651]],
[[42.784839], [90.116834], [67.911661]]])
b0 = np.asarray([[[0.394134], [0.305481], [-0.111563], [0.464680],
[-0.044974], [-0.249474], [0.344747], [-0.144516],
[0.120547], [-0.120140], [0.096371], [0.073717],
[0.044348], [-0.385316], [0.416685], [0.548801],
[0.290702], [-0.170383]],
[[-0.183913], [0.881344], [0.379849], [0.967322],
[0.231840], [0.334352], [0.516944], [0.570409
], [0.156349],
[0.909234], [0.174547], [0.813966], [0.835244],
[0.761143], [1.020271], [0.431555], [0.451456],
[-0.165681]],
[[0.538315], [1.158434], [0.827443],
[1.209531], [0.873020], [1.490418], [-0.321991],
[-0.546698], [0.387941], [0.714195], [1.914184],
[1.262890], [1.031726], [0.665484], [1.111183],
[0.780210], [0.800418], [0.016235]],
[[0.598855], [1.493059],
[1.128118], [1.443648], [0.734596], [1.172370],
[-0.883454], [-4.379097], [-3.697614], [-5.167245],
[-3.755237], [1.490307], [0.865462], [1.644435],
[1.055617], [1.160021], [0.387357], [0.157360]],
[[0.198682], [0.914774],
[1.392495], [0.880891], [1.134312], [0.968554],
[-4.343983], [-7.116327], [-3.834269], [-5.174044],
[-4.161452], [0.151857], [0.922105], [1.161507],
[0.595232], [0.990366], [1.075088], [-0.258814]],
[[0.477665], [1.602952], [1.181507],
[0.665496], [1.279437], [1.439306], [-3.930717],
[-2.732151], [0.424828], [-3.961826], [-3.768221],
[-0.472824], [1.042067], [0.807338], [1.711559],
[0.983939], [0.288783], [-0.418911]],
[[0.766438], [1.244276],
[1.286307], [0.790951], [0.995913], [0.872906],
[-3.827127], [-3.361843], [-1.077353], [-4.582801],
[-2.303175], [0.031738], [1.287701], [1.044528],
[1.383836], [1.283100], [0.241565], [-0.197527]],
[[0.411086], [1.588132],
[1.143584], [1.515480], [0.525359], [1.285077],
[-4.642499], [-4.682207], [-2.799712], [-7.874153],
[-3.598426], [0.163200], [1.050818], [1.072107],
[1.647231], [1.297448], [0.472967], [-0.365622]],
[[1.140368], [1.331310], [1.039651], [1.444111],
[0.588199], [1.074860], [-2.961898], [-3.365907],
[-9.275475], [-8.768490], [-5.709736], [0.552231],
[0.806262], [0.641553], [0.859427], [1.390415],
[-0.017225], [0.395236]],
[[1.151735], [0.551786],
[1.383289], [0.987141], [1.303332], [0.891473],
[-2.649088], [-2.601597], [-4.444745], [-6.328936],
[-3.561121], [0.097629], [0.956537], [1.460169],
[1.674805], [1.392249], [0.491228], [-0.388936]],
[[0.351013], [1.419691], [1.283733], [1.161723],
[1.707191], [0.643386], [1.431694], [0.691841],
[-1.213639], [-7.302778], [-3.354404], [-0.886146],
[0.715025], [0.994340], [0.875995], [1.663526],
[0.948911], [0.567744]],
[[0.857319], [1.086111], [0.814578], [0.970882],
[1.697328], [1.316010], [1.051087], [0.414327],
[-5.256378], [-5.877422], [-0.179250], [0.544970],
[1.264428], [0.604863], [0.999309], [1.183568],
[0.366828], [0.408546]],
[[0.746337], [1.342136], [1.351453], [1.447988],
[1.473613], [1.172948], [1.625175], [-1.097880],
[-7.496154], [-3.486610], [-0.854779], [0.951540],
[1.028474], [0.571090], [0.728899], [1.214444],
[0.993936], [0.392067]],
[[0.834584], [0.852879], [0.819633], [1.100310],
[1.119684], [0.838549], [0.308638], [-4.381538],
[-5.573656], [-0.240430], [0.531746], [1.503448],
[0.747390], [1.253126], [0.905447], [1.262200],
[0.442319], [0.442514]],
[[0.743176], [0.749403], [1.527007], [1.030668],
[1.659595], [0.818656], [0.471315], [-6.560603],
[-4.153131], [-0.389991], [0.951427], [1.292622],
[1.227982], [0.787592], [1.083349], [1.171198],
[0.320301], [0.145997]],
[[0.336351], [0.971745], [0.870700], [1.566745],
[0.849081], [1.463097], [-2.652546], [-2.183096],
[-0.495741], [-0.188252], [0.936387], [0.429778],
[1.737350], [1.646101], [0.694978], [1.069849],
[0.892659], [-0.138695]],
[[0.193120], [0.772601], [0.665977], [0.959221],
[1.782215], [1.507661], [1.003219], [1.342632
], [0.805974],
[1.522182], [0.674850], [0.820031], [0.603524
], [0.752671],
[0.938834], [1.467478], [0.745415], [0.440291]],
[[0.452432], [0.504461], [0.544565], [0.432389],
[0.740133], [0.067003], [0.893957], [0.292302
], [0.586535],
[0.411866], [0.240025], [0.279950], [0.342256],
[0.439238], [0.082237], [0.079462], [0.350406],
[-0.376255]]])
W.append(w0)
B.append(b0)
# 第2层
w1 = np.asarray([[[0.496053], [0.142468], [0.692607]],
[[-0.152569], [0.746855], [0.665359]],
[[-0.078141], [0.187796], [0.004870]]])
b1 = np.asarray([[[0.738316], [0.558320], [0.616535], [0.563276],
[0.397948], [0.403014], [0.623214], [0.648124],
[0.499537], [0.477057], [0.458188], [0.620421],
[0.698390], [0.680706], [0.483114], [0.579076]],
[[0.596898], [0.404650], [0.519849], [0.451572],
[0.522308], [0.449322], [0.525448], [0.713798],
[0.439633], [0.500006], [0.759672], [0.725525],
[0.695965], [0.449514], [0.472447], [0.482284]],
[[0.523954], [0.755020], [0.640807], [0.670937],
[0.458754], [0.624749], [-4.708748], [-4.716255],
[-4.719785], [-4.739897], [0.416815], [0.563930],
[0.471036], [0.385172], [0.570061], [0.460100]],
[[0.488801], [0.445694], [0.726974], [0.642160],
[0.389185], [0.542930], [-4.720975], [0.494188],
[0.493539], [-4.720184], [0.369179], [0.534885],
[0.693592], [0.757328], [0.404759], [0.510594]],
[[0.387318], [0.776344], [0.697167], [0.591275],
[0.678695], [0.379232], [-4.735606], [0.372957],
[0.629077], [-4.710119], [0.417072], [0.458907],
[0.603948], [0.368090], [0.548514], [0.777818]],
[[0.386241], [0.504484], [0.492332], [0.515327],
[0.734786], [0.410316], [-4.724017], [0.653110],
[0.631986], [-4.728177], [0.658651], [0.705176],
[0.620827], [0.670017], [0.385353], [0.462904]],
[[0.406522], [0.407609], [0.434104], [0.548003],
[0.671716], [0.470964], [-4.705063], [0.561526],
[-4.714012], [-4.730049], [0.713027], [0.415023],
[0.527258], [0.653842], [0.641143], [0.730861]],
[[0.399245], [0.624423], [0.586776], [0.640310],
[0.531442], [0.522265], [-4.731969], [-4.713455],
[-4.737818], [-4.726882], [0.520856], [0.613984],
[0.503019], [0.548040], [0.382262], [0.712375]],
[[0.604145], [0.507184], [0.657208], [0.454861],
[0.526420], [0.715608], [0.425322], [0.525826],
[0.519881], [-4.711324], [0.763128], [0.420057],
[0.740006], [0.488205], [0.443485], [0.460116]],
[[0.546870], [0.772369], [0.414549], [0.415155],
[0.480857], [0.598054], [0.432484], [0.766837],
[-4.708531], [-4.714308], [0.559024], [0.703133],
[0.608846], [0.459609], [0.689914], [0.389198]],
[[0.422926], [0.502196], [0.720415], [0.539214],
[0.635489], [0.552064], [0.608547], [0.535145],
[-4.723283], [0.467894], [0.476116], [0.527392],
[0.498777], [0.663084], [0.404508], [0.373993]],
[[0.425400], [0.474075], [0.628678], [0.415942],
[0.379077], [0.401436], [0.551009], [-4.741356],
[-4.714145], [0.649679], [0.422052], [0.414361],
[0.758212], [0.771625], [0.563505], [0.599782]],
[[0.534033], [0.497468], [0.427890],
[0.373989], [0.604646], [0.600752], [0.368556],
[-4.724289], [0.513206], [0.409971], [0.700321],
[0.451720], [0.419490], [0.556964], [0.546020],
[0.482902]],
[[0.392839], [0.550031], [0.598946],
[0.515958], [0.419458], [0.400969], [-4.713971],
[-4.738337], [0.695640], [0.406998], [0.538457],
[0.599610], [0.416045], [0.520507], [0.599289],
[0.683523]],
[[0.586383], [0.436032], [0.651763], [0.542701],
[0.438006], [0.594131], [0.400705], [0.745671],
[0.656867], [0.559175], [0.689486], [0.520835],
[0.385378], [0.659122], [0.651156], [0.470575]],
[[0.405100], [0.716354], [0.549392], [0.384638],
[0.413940], [0.525830], [0.380278], [0.649017],
[0.428349], [0.389288], [0.761551], [0.657696],
[0.416118], [0.655166], [0.488445], [0.488434]]])
W.append(w1)
B.append(b1)
''' W, B 参数 end '''
# 激活函数
activation = Sigmoid()
# activation = ReLU()
# 最后一跳激活函数
last_hop_activation = LastHopActivation()
# 损失函数
loss = MSELoss()
# 构建卷积对象
cvl = Convolution()
# 构建卷积神经网络对象
cnn = CVLFNN(cvl, activation, last_hop_activation, loss)
# 参数赋值
cnn.stub_set_para(0, None, W, B, activation)
# 预测
py_list = cnn.predict(sx_list, sy_list)
# 将预测结果显示如初
py = py_list[0]
py = my_image.normalize(py, my_image.NormalizationType.REV_NORMAL)
py = my_image.array_3_2(py)
messagebox.showinfo("提示", "预测结果")
show_data(py, ImageDataType.GRAY)
def test_cvl_nn():
# 1. 构建训练输入样本/输出样本
sx_list = list() # 输入样本列表
sy_list = list() # 输出样本列表
for i in range(0, 10):
# 1.1 输入样本
# 图像文件名
file_name = "../picture/number/"
file_name += "%d_sx.bmp" % i
# 取灰度值
data, err = gray_file(file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
gray = my_image.array_3_2(data)
# show_data(gray, ImageDataType.GRAY)
# 归一化
sx = my_image.normalize(data, my_image.NormalizationType.NORMAL)
# 显示归一化灰度图像
gray = my_image.array_3_2(sx)
# show_data(gray, ImageDataType.GRAY)
# 加入训练输入样本列表
sx_list.append(sx)
# 1.2 输出样本
# 图像文件名
file_name = "../picture/number/"
file_name += "%d_sy_2.bmp" % i
# 取灰度值
data, err = gray_file(file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
gray = my_image.array_3_2(data)
# show_data(gray, ImageDataType.GRAY)
# 归一化
sy = my_image.normalize(data, my_image.NormalizationType.NORMAL)
# 显示归一化灰度图像
gray = my_image.array_3_2(sy)
# show_data(gray, ImageDataType.GRAY)
# 加入训练输出样本列表
sy_list.append(sy)
# 2. 卷积神经网络的基本参数
# 每一层网络的神经元个数(这个参数，对于卷积网络而言，没意义)
neuron_count_list = None
# 最大循环训练次数
loop_max = 1
# 学习效率
rate = 0.01
# 卷积核数组大小
w_shape_list = list()
w1_shape = [3, 3, 1]
w2_shape = [3, 3, 1]
w_shape_list.append(w1_shape)
w_shape_list.append(w2_shape)
# 激活函数
activation = Sigmoid()
# activation = ReLU()
# 最后一跳激活函数
last_hop_activation = LastHopActivation()
# 损失函数
loss = MSELoss()
# 构建卷积对象
cvl = Convolution()
# 构建卷积神经网络对象
cnn = CVLFNN(cvl, activation, last_hop_activation, loss)
# 3. 训练
cnn.train(sx_list, sy_list, loop_max, neuron_count_list, rate,
w_shape_list)
# 4. 预测
py_list = cnn.predict(sx_list, sy_list)
# 将预测结果显示如初
count = len(py_list)
for i in range(0, count):
py = py_list[i]
py = my_image.normalize(py, my_image.NormalizationType.REV_NORMAL)
py = my_image.array_3_2(py)
show_data(py, ImageDataType.GRAY)
def test_cvl_nn_2():
# 1. 构建训练输入样本/输出样本
train_sx_list = list() # 训练样本列表（输入）
train_sy_list = list() # 训练样本列表（输出）
test_sx_list = list() # 测试样本列表（输入）
test_sy_list = list() # 测试样本列表（输入）
max_count = 1000
image_path = "../picture/number_cnn/"
for count in range(0, max_count):
for number in range(0, 2):
# 1.1 输入样本
# 图像路径
path = image_path + "/" + str(number)
# 文件名
sx_file_name = path + "/" + "sx_" + str(number) + "_" + str(
count) + ".bmp"
# 取灰度值
data, err = gray_file(sx_file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
# gray = my_image.array_3_2(data)
# show_data(gray, ImageDataType.GRAY)
# 归一化
sx = my_image.normalize(data, my_image.NormalizationType.NORMAL)
sx = sx / 40
# 显示归一化灰度图像
# gray = my_image.array_3_2(sx)
# show_data(gray, ImageDataType.GRAY)
# 加入训练样本列表（输入）
if count < (max_count - 1):
train_sx_list.append(sx)
# 加入测试样本列表（输入）
else:
test_sx_list.append(sx)
# 1.2 输出样本
# sy 文件名
sy_file_name = path + "/" + "sy_" + str(number) + ".bmp"
# 取灰度值
data, err = gray_file(sy_file_name, ArrayDim.THREE)
# 将图像数据中的0转换为极小值
my_image.array_0_tiny(data)
# 显示灰度图像
# gray = my_image.array_3_2(data)
# show_data(gray, ImageDataType.GRAY)
# 归一化
sy = my_image.normalize(data, my_image.NormalizationType.NORMAL)
# 显示归一化灰度图像
# gray = my_image.array_3_2(sy)
# show_data(gray, ImageDataType.GRAY)
# 加入训练样本列表（输出）
if count < (max_count - 1):
train_sy_list.append(sy)
# 加入训练样本列表（输出）
else:
test_sy_list.append(sy)
# 2. 卷积神经网络的基本参数
# 每一层网络的神经元个数(这个参数，对于卷积网络而言，没意义)
neuron_count_list = None
# 最大循环训练次数
loop_max = 1
# 学习效率
rate = 0.01
# 卷积核数组大小
w_shape_list = list()
w1_shape = [3, 3, 1]
w2_shape = [3, 3, 1]
w_shape_list.append(w1_shape)
w_shape_list.append(w2_shape)
# 激活函数
activation = Sigmoid()
# activation = ReLU()
# 最后一跳激活函数
last_hop_activation = LastHopActivation()
# 损失函数
loss = MSELoss()
# 构建卷积对象
cvl = Convolution()
# 构建卷积神经网络对象
cnn = CVLFNN(cvl, activation, last_hop_activation, loss)
# 3. 训练
# train_sx_group = get_sample_group(train_sx_list)
# train_sy_group = get_sample_group(train_sy_list)
# cnn.train(train_sx_group, train_sy_group, loop_max, neuron_count_list, rate, w_shape_list)
cnn.train(train_sx_list, train_sy_list, loop_max, neuron_count_list, rate,
w_shape_list)
# 4. 预测
py_list = cnn.predict(test_sx_list, test_sy_list)
# 将预测结果显示如初
count = len(py_list)
for number in range(0, count):
py = py_list[number]
py = my_image.normalize(py, my_image.NormalizationType.REV_NORMAL)
py = my_image.array_3_2(py)
show_data(py, ImageDataType.GRAY)