任務(wù)描述：

本次實(shí)踐是一個(gè)多分類任務(wù)，需要將照片中的每個(gè)字符分別進(jìn)行識(shí)別，完成車牌識(shí)別

實(shí)踐平臺(tái)：百度AI實(shí)訓(xùn)平臺(tái)-AI Studio、PaddlePaddle1.8.0 動(dòng)態(tài)圖

深度神經(jīng)網(wǎng)絡(luò)(DNN)

深度神經(jīng)網(wǎng)絡(luò)（Deep Neural Networks，簡(jiǎn)稱DNN）是深度學(xué)習(xí)的基礎(chǔ)，其結(jié)構(gòu)為input、hidden（可有多層）、output，每層均為全連接。

數(shù)據(jù)集介紹

• 數(shù)據(jù)集文件名為characterData.zip，其中有65個(gè)文件夾

• 包含0-9，A-Z，以及各省簡(jiǎn)稱

• 圖片為12020的灰度圖像

• 本次實(shí)驗(yàn)中，取其中的10%作為測(cè)試集，90%作為訓(xùn)練集

# 查看當(dāng)前掛載的數(shù)據(jù)集目錄, 該目錄下的變更重啟環(huán)境后會(huì)自動(dòng)還原 # View dataset directory. This directory will be recovered automatically after resetting environment. !ls /home/aistudio/data data47142 #導(dǎo)入需要的包 import os import zipfile import random import json import cv2 import numpy as np from PIL import Image import paddle import paddle.fluid as fluid from paddle.fluid.dygraph import Linear import matplotlib.pyplot as plt from paddle.fluid.dygraph import Linear,Conv2D,Pool2D 2020-07-29 21:04:07,361-INFO: font search path ['/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/ttf', '/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/afm', '/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/pdfcorefonts'] 2020-07-29 21:04:07,788-INFO: generated new fontManager

1、數(shù)據(jù)準(zhǔn)備

''' 參數(shù)配置 ''' train_parameters = {"input_size": [1, 20, 20], #輸入圖片的shape"class_dim": -1, #分類數(shù)"src_path":"data/data47142/characterData.zip", #原始數(shù)據(jù)集路徑"target_path":"/home/aistudio/data/dataset", #要解壓的路徑 "train_list_path": "./train_data.txt", #train_data.txt路徑"eval_list_path": "./val_data.txt", #eval_data.txt路徑"label_dict":{}, #標(biāo)簽字典"readme_path": "/home/aistudio/data/readme.json", #readme.json路徑"num_epochs": 20, #訓(xùn)練輪數(shù)"train_batch_size": 32, #批次的大小"learning_strategy": { #優(yōu)化函數(shù)相關(guān)的配置"lr": 0.001 #超參數(shù)學(xué)習(xí)率} } def unzip_data(src_path,target_path):'''解壓原始數(shù)據(jù)集，將src_path路徑下的zip包解壓至data/dataset目錄下'''if(not os.path.isdir(target_path)): z = zipfile.ZipFile(src_path, 'r')z.extractall(path=target_path)z.close()else:print("文件已解壓") def get_data_list(target_path,train_list_path,eval_list_path):'''生成數(shù)據(jù)列表'''#存放所有類別的信息class_detail = []#獲取所有類別保存的文件夾名稱data_list_path=target_pathclass_dirs = os.listdir(data_list_path)if '__MACOSX' in class_dirs:class_dirs.remove('__MACOSX')# #總的圖像數(shù)量all_class_images = 0# #存放類別標(biāo)簽class_label=0# #存放類別數(shù)目class_dim = 0# #存儲(chǔ)要寫(xiě)進(jìn)eval.txt和train.txt中的內(nèi)容trainer_list=[]eval_list=[]#讀取每個(gè)類別for class_dir in class_dirs:if class_dir != ".DS_Store":class_dim += 1#每個(gè)類別的信息class_detail_list = {}eval_sum = 0trainer_sum = 0#統(tǒng)計(jì)每個(gè)類別有多少?gòu)垐D片class_sum = 0#獲取類別路徑 path = os.path.join(data_list_path,class_dir)# print(path)# 獲取所有圖片img_paths = os.listdir(path)for img_path in img_paths: # 遍歷文件夾下的每個(gè)圖片if img_path =='.DS_Store':continuename_path = os.path.join(path,img_path) # 每張圖片的路徑if class_sum % 10 == 0: # 每10張圖片取一個(gè)做驗(yàn)證數(shù)據(jù)eval_sum += 1 # eval_sum為測(cè)試數(shù)據(jù)的數(shù)目eval_list.append(name_path + "\t%d" % class_label + "\n")else:trainer_sum += 1 trainer_list.append(name_path + "\t%d" % class_label + "\n")#trainer_sum測(cè)試數(shù)據(jù)的數(shù)目class_sum += 1 #每類圖片的數(shù)目all_class_images += 1 #所有類圖片的數(shù)目# 說(shuō)明的json文件的class_detail數(shù)據(jù)class_detail_list['class_name'] = class_dir #類別名稱class_detail_list['class_label'] = class_label #類別標(biāo)簽class_detail_list['class_eval_images'] = eval_sum #該類數(shù)據(jù)的測(cè)試集數(shù)目class_detail_list['class_trainer_images'] = trainer_sum #該類數(shù)據(jù)的訓(xùn)練集數(shù)目class_detail.append(class_detail_list) #初始化標(biāo)簽列表train_parameters['label_dict'][str(class_label)] = class_dirclass_label += 1#初始化分類數(shù)train_parameters['class_dim'] = class_dimprint(train_parameters)#亂序 random.shuffle(eval_list)with open(eval_list_path, 'a') as f:for eval_image in eval_list:f.write(eval_image) #亂序 random.shuffle(trainer_list) with open(train_list_path, 'a') as f2:for train_image in trainer_list:f2.write(train_image) # 說(shuō)明的json文件信息readjson = {}readjson['all_class_name'] = data_list_path #文件父目錄readjson['all_class_images'] = all_class_imagesreadjson['class_detail'] = class_detailjsons = json.dumps(readjson, sort_keys=True, indent=4, separators=(',', ': '))with open(train_parameters['readme_path'],'w') as f:f.write(jsons)print ('生成數(shù)據(jù)列表完成！') def data_reader(file_list):'''自定義data_reader'''def reader():with open(file_list, 'r') as f:lines = [line.strip() for line in f]for line in lines:img_path, lab = line.strip().split('\t')img = cv2.imread(img_path)img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)img = np.array(img).astype('float32')img = img/255.0yield img, int(lab) return reader ''' 參數(shù)初始化 ''' src_path=train_parameters['src_path'] target_path=train_parameters['target_path'] train_list_path=train_parameters['train_list_path'] eval_list_path=train_parameters['eval_list_path'] batch_size=train_parameters['train_batch_size'] ''' 解壓原始數(shù)據(jù)到指定路徑 ''' unzip_data(src_path,target_path)#每次生成數(shù)據(jù)列表前，首先清空train.txt和eval.txt with open(train_list_path, 'w') as f: f.seek(0)f.truncate() with open(eval_list_path, 'w') as f: f.seek(0)f.truncate() #生成數(shù)據(jù)列表 get_data_list(target_path,train_list_path,eval_list_path)''' 構(gòu)造數(shù)據(jù)提供器 ''' train_reader = paddle.batch(data_reader(train_list_path),batch_size=batch_size,drop_last=True) eval_reader = paddle.batch(data_reader(eval_list_path),batch_size=batch_size,drop_last=True) 文件已解壓 {'input_size': [1, 20, 20], 'class_dim': 65, 'src_path': 'data/data47142/characterData.zip', 'target_path': '/home/aistudio/data/dataset', 'train_list_path': './train_data.txt', 'eval_list_path': './val_data.txt', 'label_dict': {'0': 'e1', '1': 'X', '2': 'hu', '3': 'C', '4': 'P', '5': 'shan', '6': 'su', '7': '1', '8': 'M', '9': 'zang', '10': 'ji', '11': 'wan', '12': 'gui1', '13': 'yue', '14': 'Q', '15': 'ning', '16': 'B', '17': 'jin', '18': 'liao', '19': 'min', '20': 'H', '21': 'gan1', '22': '3', '23': 'S', '24': 'K', '25': 'meng', '26': '6', '27': '7', '28': '9', '29': 'gan', '30': 'A', '31': 'V', '32': 'D', '33': 'Y', '34': 'N', '35': 'F', '36': 'lu', '37': 'xin', '38': 'E', '39': 'G', '40': 'cuan', '41': 'yun', '42': 'jing', '43': 'T', '44': '4', '45': 'L', '46': 'sx', '47': '8', '48': 'qing', '49': 'hei', '50': '2', '51': 'yu', '52': 'R', '53': 'zhe', '54': 'qiong', '55': 'gui', '56': '0', '57': 'Z', '58': 'U', '59': 'jl', '60': 'xiang', '61': 'W', '62': 'J', '63': '5', '64': 'yu1'}, 'readme_path': '/home/aistudio/data/readme.json', 'num_epochs': 20, 'train_batch_size': 32, 'learning_strategy': {'lr': 0.001}} 生成數(shù)據(jù)列表完成！ Batch=0 Batchs=[] all_train_accs=[] def draw_train_acc(Batchs, train_accs):title="training accs"plt.title(title, fontsize=24)plt.xlabel("batch", fontsize=14)plt.ylabel("acc", fontsize=14)plt.plot(Batchs, train_accs, color='green', label='training accs')plt.legend()plt.grid()plt.show()all_train_loss=[] def draw_train_loss(Batchs, train_loss):title="training loss"plt.title(title, fontsize=24)plt.xlabel("batch", fontsize=14)plt.ylabel("loss", fontsize=14)plt.plot(Batchs, train_loss, color='red', label='training loss')plt.legend()plt.grid()plt.show()

2、定義模型

###在以下cell中完成DNN網(wǎng)絡(luò)的定義###

#定義DNN網(wǎng)絡(luò) class MyDNN(fluid.dygraph.Layer):'''DNN網(wǎng)絡(luò)'''def __init__(self):super(MyDNN,self).__init__()self.hidden1_1 = Conv2D(1,28,5,1) #通道數(shù)、卷積核個(gè)數(shù)、卷積核大小self.hidden1_2 = Pool2D(pool_size=2,pool_type='max',pool_stride=1)self.hidden2_1 = Conv2D(28,32,3,1)self.hidden2_2 = Pool2D(pool_size=2,pool_type='max',pool_stride=1)self.hidden3 = Conv2D(32,32,3,1)self.hidden4 = Linear(32*10*10,65,act='softmax')def forward(self,input): # forward 定義執(zhí)行實(shí)際運(yùn)行時(shí)網(wǎng)絡(luò)的執(zhí)行邏輯'''前向計(jì)算'''#print(input.shape)x = self.hidden1_1(input)#print(x.shape)x = self.hidden1_2(x)#print(x.shape)x = self.hidden2_1(x)#print(x.shape)x = self.hidden2_2(x)#print(x.shape)x = self.hidden3(x)#print(x.shape)x = fluid.layers.reshape(x, shape=[-1, 32*10*10])y = self.hidden4(x)return y

3、訓(xùn)練模型

with fluid.dygraph.guard():model=MyDNN() #模型實(shí)例化model.train() #訓(xùn)練模式opt=fluid.optimizer.SGDOptimizer(learning_rate=train_parameters['learning_strategy']['lr'], parameter_list=model.parameters())#優(yōu)化器選用SGD隨機(jī)梯度下降，學(xué)習(xí)率為0.001.epochs_num=train_parameters['num_epochs'] #迭代次數(shù)for pass_num in range(epochs_num):for batch_id,data in enumerate(train_reader()):images=np.array([x[0].reshape(1,20,20) for x in data],np.float32)labels = np.array([x[1] for x in data]).astype('int64')labels = labels[:, np.newaxis]image=fluid.dygraph.to_variable(images)label=fluid.dygraph.to_variable(labels)predict=model(image) #數(shù)據(jù)傳入modelloss=fluid.layers.cross_entropy(predict,label)avg_loss=fluid.layers.mean(loss)#獲取loss值acc=fluid.layers.accuracy(predict,label)#計(jì)算精度if batch_id!=0 and batch_id%50==0:Batch = Batch+50 Batchs.append(Batch)all_train_loss.append(avg_loss.numpy()[0])all_train_accs.append(acc.numpy()[0])print("train_pass:{},batch_id:{},train_loss:{},train_acc:{}".format(pass_num,batch_id,avg_loss.numpy(),acc.numpy()))avg_loss.backward() opt.minimize(avg_loss) #優(yōu)化器對(duì)象的minimize方法對(duì)參數(shù)進(jìn)行更新 model.clear_gradients() #model.clear_gradients()來(lái)重置梯度fluid.save_dygraph(model.state_dict(),'MyDNN')#保存模型draw_train_acc(Batchs,all_train_accs) draw_train_loss(Batchs,all_train_loss) train_pass:0,batch_id:50,train_loss:[2.5459664],train_acc:[0.375] train_pass:0,batch_id:100,train_loss:[2.475458],train_acc:[0.46875] train_pass:0,batch_id:150,train_loss:[1.2719567],train_acc:[0.6875] train_pass:0,batch_id:200,train_loss:[1.4321089],train_acc:[0.6875] train_pass:0,batch_id:250,train_loss:[0.69725776],train_acc:[0.875] train_pass:0,batch_id:300,train_loss:[1.1576208],train_acc:[0.75] train_pass:0,batch_id:350,train_loss:[1.207567],train_acc:[0.75] train_pass:0,batch_id:400,train_loss:[0.6369293],train_acc:[0.84375] train_pass:0,batch_id:450,train_loss:[0.99959254],train_acc:[0.78125]

4、模型評(píng)估

#模型評(píng)估 with fluid.dygraph.guard():accs = []model_dict, _ = fluid.load_dygraph('MyDNN')model = MyDNN()model.load_dict(model_dict) #加載模型參數(shù)model.eval() #訓(xùn)練模式for batch_id,data in enumerate(eval_reader()):#測(cè)試集images=np.array([x[0].reshape(1,20,20) for x in data],np.float32)labels = np.array([x[1] for x in data]).astype('int64')labels = labels[:, np.newaxis]image=fluid.dygraph.to_variable(images)label=fluid.dygraph.to_variable(labels) predict=model(image) acc=fluid.layers.accuracy(predict,label)accs.append(acc.numpy()[0])avg_acc = np.mean(accs)print(avg_acc) 0.96568626

5、使用模型

5.1對(duì)車牌圖像進(jìn)行預(yù)處理

# 對(duì)車牌圖片進(jìn)行處理，分割出車牌中的每一個(gè)字符并保存 license_plate = cv2.imread('work/車牌.png') gray_plate = cv2.cvtColor(license_plate, cv2.COLOR_RGB2GRAY) ret, binary_plate = cv2.threshold(gray_plate, 175, 255, cv2.THRESH_BINARY) #ret：閾值，binary_plate：根據(jù)閾值處理后的圖像數(shù)據(jù) # 按列統(tǒng)計(jì)像素分布 result = [] for col in range(binary_plate.shape[1]):result.append(0)for row in range(binary_plate.shape[0]):result[col] = result[col] + binary_plate[row][col]/255 # print(result) #記錄車牌中字符的位置 character_dict = {} num = 0 i = 0 while i < len(result):if result[i] == 0:i += 1else:index = i + 1while result[index] != 0:index += 1character_dict[num] = [i, index-1]num += 1i = index # print(character_dict) #將每個(gè)字符填充，并存儲(chǔ) characters = [] for i in range(8):if i==2:continuepadding = (170 - (character_dict[i][1] - character_dict[i][0])) / 2#將單個(gè)字符圖像填充為170*170ndarray = np.pad(binary_plate[:,character_dict[i][0]:character_dict[i][1]], ((0,0), (int(padding), int(padding))), 'constant', constant_values=(0,0))ndarray = cv2.resize(ndarray, (20,20))cv2.imwrite('work/' + str(i) + '.png', ndarray)characters.append(ndarray)def load_image(path):img = paddle.dataset.image.load_image(file=path, is_color=False)img = img.astype('float32')img = img[np.newaxis, ] / 255.0return img

5.2 對(duì)標(biāo)簽進(jìn)行轉(zhuǎn)換

#將標(biāo)簽進(jìn)行轉(zhuǎn)換 print('Label:',train_parameters['label_dict']) match = {'A':'A','B':'B','C':'C','D':'D','E':'E','F':'F','G':'G','H':'H','I':'I','J':'J','K':'K','L':'L','M':'M','N':'N','O':'O','P':'P','Q':'Q','R':'R','S':'S','T':'T','U':'U','V':'V','W':'W','X':'X','Y':'Y','Z':'Z','yun':'云','cuan':'川','hei':'黑','zhe':'浙','ning':'寧','jin':'津','gan':'贛','hu':'滬','liao':'遼','jl':'吉','qing':'青','zang':'藏','e1':'鄂','meng':'蒙','gan1':'甘','qiong':'瓊','shan':'陜','min':'閩','su':'蘇','xin':'新','wan':'皖','jing':'京','xiang':'湘','gui':'貴','yu1':'渝','yu':'豫','ji':'冀','yue':'粵','gui1':'桂','sx':'晉','lu':'魯','0':'0','1':'1','2':'2','3':'3','4':'4','5':'5','6':'6','7':'7','8':'8','9':'9'} L = 0 LABEL ={} for V in train_parameters['label_dict'].values():LABEL[str(L)] = match[V]L += 1 print(LABEL) Label: {'0': 'Y', '1': 'hu', '2': '2', '3': 'cuan', '4': 'ji', '5': 'T', '6': '5', '7': '9', '8': 'yue', '9': 'yun', '10': 'hei', '11': 'H', '12': 'G', '13': 'L', '14': 'gan1', '15': '1', '16': '6', '17': 'qing', '18': 'K', '19': 'gui', '20': 'C', '21': 'gan', '22': 'yu', '23': 'X', '24': '8', '25': 'sx', '26': 'xiang', '27': 'S', '28': 'ning', '29': 'jin', '30': 'xin', '31': 'jl', '32': 'min', '33': 'B', '34': 'J', '35': 'yu1', '36': 'N', '37': 'gui1', '38': 'M', '39': '4', '40': 'D', '41': 'R', '42': 'shan', '43': '7', '44': 'E', '45': 'Q', '46': 'A', '47': 'U', '48': '3', '49': 'F', '50': 'W', '51': 'meng', '52': 'liao', '53': 'jing', '54': 'zang', '55': 'lu', '56': 'V', '57': 'Z', '58': '0', '59': 'su', '60': 'wan', '61': 'P', '62': 'zhe', '63': 'e1', '64': 'qiong'} {'0': 'Y', '1': '滬', '2': '2', '3': '川', '4': '冀', '5': 'T', '6': '5', '7': '9', '8': '粵', '9': '云', '10': '黑', '11': 'H', '12': 'G', '13': 'L', '14': '甘', '15': '1', '16': '6', '17': '青', '18': 'K', '19': '貴', '20': 'C', '21': '贛', '22': '豫', '23': 'X', '24': '8', '25': '晉', '26': '湘', '27': 'S', '28': '寧', '29': '津', '30': '新', '31': '吉', '32': '閩', '33': 'B', '34': 'J', '35': '渝', '36': 'N', '37': '桂', '38': 'M', '39': '4', '40': 'D', '41': 'R', '42': '陜', '43': '7', '44': 'E', '45': 'Q', '46': 'A', '47': 'U', '48': '3', '49': 'F', '50': 'W', '51': '蒙', '52': '遼', '53': '京', '54': '藏', '55': '魯', '56': 'V', '57': 'Z', '58': '0', '59': '蘇', '60': '皖', '61': 'P', '62': '浙', '63': '鄂', '64': '瓊'}

5.3 使用模型進(jìn)行預(yù)測(cè)

#構(gòu)建預(yù)測(cè)動(dòng)態(tài)圖過(guò)程 with fluid.dygraph.guard():model=MyDNN()#模型實(shí)例化model_dict,_=fluid.load_dygraph('MyDNN')model.load_dict(model_dict)#加載模型參數(shù)model.eval()#評(píng)估模式lab=[]for i in range(8):if i==2:continueinfer_imgs = []infer_imgs.append(load_image('work/' + str(i) + '.png'))infer_imgs = np.array(infer_imgs)infer_imgs = fluid.dygraph.to_variable(infer_imgs)result=model(infer_imgs)lab.append(np.argmax(result.numpy())) print(lab) display(Image.open('work/車牌.png')) for i in range(len(lab)):print(LABEL[str(lab[i])],end='') [55, 46, 16, 24, 16, 44, 34] 魯A686EJ

總結(jié)

以上是生活随笔為你收集整理的百度 AI Studio 车牌识别的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。