深度神經網絡實踐：車牌識別

• 任務描述：
• 深度神經網絡(DNN)
• 數據集介紹
• 1、數據準備
• 2、定義模型
• 3、訓練模型
• 4、模型評估
• 5、使用模型
• • 5.1對車牌圖像進行預處理
  • 5.2 對標簽進行轉換
  • 5.3 使用模型進行預測

注意只有需要訓練和更新權重參數的算子才能作為單獨的一層，比如全連接層（FC）、卷積層（CONV），而激活函數（ReLU、Sigmoid等）、池化（MAX POOL、AVAGE POOL等）則不能算做一層。

任務描述：

本次實踐是一個多分類任務，需要將照片中的每個字符分別進行識別，完成車牌的識別

實踐平臺：百度AI實訓平臺-AI Studio、PaddlePaddle1.8.0 動態圖

深度神經網絡(DNN)

深度神經網絡（Deep Neural Networks，簡稱DNN）是深度學習的基礎，其結構為input、hidden（可有多層）、output，每層均為全連接。

數據集介紹

• 數據集文件名為characterData.zip，其中有65個文件夾

• 包含0-9，A-Z，以及各省簡稱

• 圖片為12020的灰度圖像

• 本次實驗中，取其中的10%作為測試集，90%作為訓練集

#導入需要的包 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

1、數據準備

''' 參數配置 ''' train_parameters = {"input_size": [1, 20, 20], #輸入圖片的shape"class_dim": -1, #分類數"src_path":"data/data23617/characterData.zip", #原始數據集路徑"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":{}, #標簽字典"readme_path": "/home/aistudio/data/readme.json", #readme.json路徑"num_epochs": 1, #訓練輪數"train_batch_size": 32, #批次的大小"learning_strategy": { #優化函數相關的配置"lr": 0.001 #超參數學習率} } def unzip_data(src_path,target_path):'''解壓原始數據集，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):'''生成數據列表'''#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')# #總的圖像數量all_class_images = 0# #存放類別標簽class_label=0# #存放類別數目class_dim = 0# #存儲要寫進eval.txt和train.txt中的內容trainer_list=[]eval_list=[]#讀取每個類別for class_dir in class_dirs:if class_dir != ".DS_Store":class_dim += 1#每個類別的信息class_detail_list = {}eval_sum = 0trainer_sum = 0#統計每個類別有多少張圖片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張圖片取一個做驗證數據eval_sum += 1 # eval_sum為測試數據的數目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測試數據的數目class_sum += 1 #每類圖片的數目all_class_images += 1 #所有類圖片的數目# 說明的json文件的class_detail數據class_detail_list['class_name'] = class_dir #類別名稱class_detail_list['class_label'] = class_label #類別標簽class_detail_list['class_eval_images'] = eval_sum #該類數據的測試集數目class_detail_list['class_trainer_images'] = trainer_sum #該類數據的訓練集數目class_detail.append(class_detail_list) #初始化標簽列表train_parameters['label_dict'][str(class_label)] = class_dirclass_label += 1#初始化分類數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 ('生成數據列表完成！') 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'] ''' 解壓原始數據到指定路徑 ''' unzip_data(src_path,target_path)#每次生成數據列表前，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() #生成數據列表 get_data_list(target_path,train_list_path,eval_list_path)''' 構造數據提供器 ''' 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) 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

#定義DNN網絡 class MyDNN(fluid.dygraph.Layer):def __init__(self):super(MyDNN,self).__init__()self.hidden1 = Linear(20*20,200,act='relu')self.hidden2 = Linear(200,100,act='relu')self.hidden3 = Linear(100,100,act='relu')self.out = Linear(100,65,act='softmax')def forward(self,input): # forward 定義執行實際運行時網絡的執行邏輯x = fluid.layers.reshape(input, shape=[-1,20*20]) #-1 表示這個維度的值是從x的元素總數和剩余維度推斷出來的，有且只能有一個維度設置為-1# print(x.shape)x = self.hidden1(x)# print('1', x.shape)x = self.hidden2(x)# print('2',x.shape)x = self.hidden3(x)# print('3',x.shape)y = self.out(x)# print('4',y.shape)return y

3、訓練模型

with fluid.dygraph.guard():model=MyDNN() #模型實例化model.train() #訓練模式opt=fluid.optimizer.SGDOptimizer(learning_rate=train_parameters['learning_strategy']['lr'], parameter_list=model.parameters())#優化器選用SGD隨機梯度下降，學習率為0.001.epochs_num=train_parameters['num_epochs'] #迭代次數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) #數據傳入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) #優化器對象的minimize方法對參數進行更新 model.clear_gradients() #model.clear_gradients()來重置梯度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)

4、模型評估

#模型評估 with fluid.dygraph.guard():accs = []model_dict, _ = fluid.load_dygraph('MyDNN')model = MyDNN()model.load_dict(model_dict) #加載模型參數model.eval() #訓練模式for batch_id,data in enumerate(eval_reader()):#測試集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) 0.082720585

5、使用模型

5.1對車牌圖像進行預處理

# 對車牌圖片進行處理，分割出車牌中的每一個字符并保存 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：根據閾值處理后的圖像數據 # 按列統計像素分布 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) #將每個字符填充，并存儲 characters = [] for i in range(8):if i==2:continuepadding = (170 - (character_dict[i][1] - character_dict[i][0])) / 2#將單個字符圖像填充為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 對標簽進行轉換

#將標簽進行轉換 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)

5.3 使用模型進行預測

#構建預測動態圖過程 with fluid.dygraph.guard():model=MyDNN()#模型實例化model_dict,_=fluid.load_dygraph('MyDNN')model.load_dict(model_dict)#加載模型參數model.eval()#評估模式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='')

總結

以上是生活随笔為你收集整理的深度神经网络实践：车牌识别的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。