文章詳情頁

用python搭建一個(gè)花卉識(shí)別系統(tǒng)

瀏覽：2日期：2022-06-16 11:17:00

目錄一.開源神經(jīng)網(wǎng)絡(luò)(AlexNet)1.獲取數(shù)據(jù)集2.神經(jīng)網(wǎng)絡(luò)模型3.訓(xùn)練神經(jīng)網(wǎng)絡(luò)4.對(duì)模型進(jìn)行預(yù)測(cè)二、花卉識(shí)別系統(tǒng)搭建（flask）1.構(gòu)建頁面：2.調(diào)用神經(jīng)網(wǎng)絡(luò)模型3.系統(tǒng)識(shí)別結(jié)果4.啟動(dòng)系統(tǒng)：三、總結(jié)一.開源神經(jīng)網(wǎng)絡(luò)(AlexNet)1.獲取數(shù)據(jù)集

使用步驟如下：* （1）在data_set文件夾下創(chuàng)建新文件夾'flower_data'* （2）點(diǎn)擊鏈接下載花分類數(shù)據(jù)集download.tensorflow.org/example_im…* （3）解壓數(shù)據(jù)集到flower_data文件夾下* （4）執(zhí)行'split_data.py'腳本自動(dòng)將數(shù)據(jù)集劃分成訓(xùn)練集train和驗(yàn)證集val

split_data.py

import osfrom shutil import copy, rmtreeimport random def mk_file(file_path: str): if os.path.exists(file_path):# 如果文件夾存在，則先刪除原文件夾在重新創(chuàng)建rmtree(file_path) os.makedirs(file_path) def main(): # 保證隨機(jī)可復(fù)現(xiàn) random.seed(0) # 將數(shù)據(jù)集中10%的數(shù)據(jù)劃分到驗(yàn)證集中 split_rate = 0.1 # 指向你解壓后的flower_photos文件夾 cwd = os.getcwd() data_root = os.path.join(cwd, 'flower_data') origin_flower_path = os.path.join(data_root, 'flower_photos') assert os.path.exists(origin_flower_path) flower_class = [cla for cla in os.listdir(origin_flower_path) if os.path.isdir(os.path.join(origin_flower_path, cla))] # 建立保存訓(xùn)練集的文件夾 train_root = os.path.join(data_root, 'train') mk_file(train_root) for cla in flower_class:# 建立每個(gè)類別對(duì)應(yīng)的文件夾mk_file(os.path.join(train_root, cla)) # 建立保存驗(yàn)證集的文件夾 val_root = os.path.join(data_root, 'val') mk_file(val_root) for cla in flower_class:# 建立每個(gè)類別對(duì)應(yīng)的文件夾mk_file(os.path.join(val_root, cla)) for cla in flower_class:cla_path = os.path.join(origin_flower_path, cla)images = os.listdir(cla_path)num = len(images)# 隨機(jī)采樣驗(yàn)證集的索引eval_index = random.sample(images, k=int(num*split_rate))for index, image in enumerate(images): if image in eval_index:# 將分配至驗(yàn)證集中的文件復(fù)制到相應(yīng)目錄image_path = os.path.join(cla_path, image)new_path = os.path.join(val_root, cla)copy(image_path, new_path) else:# 將分配至訓(xùn)練集中的文件復(fù)制到相應(yīng)目錄image_path = os.path.join(cla_path, image)new_path = os.path.join(train_root, cla)copy(image_path, new_path) print('r[{}] processing [{}/{}]'.format(cla, index+1, num), end='') # processing barprint() print('processing done!') if __name__ == ’__main__’: main()2.神經(jīng)網(wǎng)絡(luò)模型

model.py

import torch.nn as nnimport torch class AlexNet(nn.Module): def __init__(self, num_classes=1000, init_weights=False):super(AlexNet, self).__init__()# 用nn.Sequential()將網(wǎng)絡(luò)打包成一個(gè)模塊，精簡(jiǎn)代碼self.features = nn.Sequential( # 卷積層提取圖像特征 nn.Conv2d(3, 48, kernel_size=11, stride=4, padding=2), # input[3, 224, 224] output[48, 55, 55] nn.ReLU(inplace=True), # 直接修改覆蓋原值，節(jié)省運(yùn)算內(nèi)存 nn.MaxPool2d(kernel_size=3, stride=2), # output[48, 27, 27] nn.Conv2d(48, 128, kernel_size=5, padding=2), # output[128, 27, 27] nn.ReLU(inplace=True), nn.MaxPool2d(kernel_size=3, stride=2), # output[128, 13, 13] nn.Conv2d(128, 192, kernel_size=3, padding=1), # output[192, 13, 13] nn.ReLU(inplace=True), nn.Conv2d(192, 192, kernel_size=3, padding=1), # output[192, 13, 13] nn.ReLU(inplace=True), nn.Conv2d(192, 128, kernel_size=3, padding=1), # output[128, 13, 13] nn.ReLU(inplace=True), nn.MaxPool2d(kernel_size=3, stride=2), # output[128, 6, 6])self.classifier = nn.Sequential( # 全連接層對(duì)圖像分類 nn.Dropout(p=0.5), # Dropout 隨機(jī)失活神經(jīng)元，默認(rèn)比例為0.5 nn.Linear(128 * 6 * 6, 2048), nn.ReLU(inplace=True), nn.Dropout(p=0.5), nn.Linear(2048, 2048), nn.ReLU(inplace=True), nn.Linear(2048, num_classes),)if init_weights: self._initialize_weights() # 前向傳播過程 def forward(self, x):x = self.features(x)x = torch.flatten(x, start_dim=1)# 展平后再傳入全連接層x = self.classifier(x)return x# 網(wǎng)絡(luò)權(quán)重初始化，實(shí)際上 pytorch 在構(gòu)建網(wǎng)絡(luò)時(shí)會(huì)自動(dòng)初始化權(quán)重 def _initialize_weights(self):for m in self.modules(): if isinstance(m, nn.Conv2d): # 若是卷積層nn.init.kaiming_normal_(m.weight, mode=’fan_out’, # 用（何）kaiming_normal_法初始化權(quán)重nonlinearity=’relu’)if m.bias is not None: nn.init.constant_(m.bias, 0) # 初始化偏重為0 elif isinstance(m, nn.Linear): # 若是全連接層nn.init.normal_(m.weight, 0, 0.01) # 正態(tài)分布初始化nn.init.constant_(m.bias, 0) # 初始化偏重為03.訓(xùn)練神經(jīng)網(wǎng)絡(luò)

train.py

# 導(dǎo)入包import torchimport torch.nn as nnfrom torchvision import transforms, datasets, utilsimport matplotlib.pyplot as pltimport numpy as npimport torch.optim as optimfrom model import AlexNetimport osimport jsonimport time # 使用GPU訓(xùn)練device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')with open(os.path.join('train.log'), 'a') as log: log.write(str(device)+'n') #數(shù)據(jù)預(yù)處理data_transform = { 'train': transforms.Compose([transforms.RandomResizedCrop(224), # 隨機(jī)裁剪，再縮放成 224×224 transforms.RandomHorizontalFlip(p=0.5), # 水平方向隨機(jī)翻轉(zhuǎn)，概率為 0.5, 即一半的概率翻轉(zhuǎn), 一半的概率不翻轉(zhuǎn) transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]), 'val': transforms.Compose([transforms.Resize((224, 224)), # cannot 224, must (224, 224) transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])} #導(dǎo)入、加載訓(xùn)練集# 導(dǎo)入訓(xùn)練集#train_set = torchvision.datasets.CIFAR10(root=’./data’, # 數(shù)據(jù)集存放目錄# train=True, # 表示是數(shù)據(jù)集中的訓(xùn)練集#download=True, # 第一次運(yùn)行時(shí)為True，下載數(shù)據(jù)集，下載完成后改為False#transform=transform) # 預(yù)處理過程# 加載訓(xùn)練集 #train_loader = torch.utils.data.DataLoader(train_set, # 導(dǎo)入的訓(xùn)練集# batch_size=50, # 每批訓(xùn)練的樣本數(shù)# shuffle=False, # 是否打亂訓(xùn)練集# num_workers=0) # num_workers在windows下設(shè)置為0 # 獲取圖像數(shù)據(jù)集的路徑data_root = os.path.abspath(os.path.join(os.getcwd(), '../..')) # get data root path 返回上上層目錄image_path = data_root + '/jqsj/data_set/flower_data/' # flower data_set path # 導(dǎo)入訓(xùn)練集并進(jìn)行預(yù)處理train_dataset = datasets.ImageFolder(root=image_path + '/train', transform=data_transform['train'])train_num = len(train_dataset) # 按batch_size分批次加載訓(xùn)練集train_loader = torch.utils.data.DataLoader(train_dataset,# 導(dǎo)入的訓(xùn)練集 batch_size=32, # 每批訓(xùn)練的樣本數(shù) shuffle=True,# 是否打亂訓(xùn)練集 num_workers=0)# 使用線程數(shù)，在windows下設(shè)置為0 #導(dǎo)入、加載驗(yàn)證集# 導(dǎo)入驗(yàn)證集并進(jìn)行預(yù)處理validate_dataset = datasets.ImageFolder(root=image_path + '/val',transform=data_transform['val'])val_num = len(validate_dataset) # 加載驗(yàn)證集validate_loader = torch.utils.data.DataLoader(validate_dataset,# 導(dǎo)入的驗(yàn)證集 batch_size=32, shuffle=True, num_workers=0) # 存儲(chǔ) 索引：標(biāo)簽的字典# 字典，類別：索引 {’daisy’:0, ’dandelion’:1, ’roses’:2, ’sunflower’:3, ’tulips’:4}flower_list = train_dataset.class_to_idx# 將 flower_list 中的 key 和 val 調(diào)換位置cla_dict = dict((val, key) for key, val in flower_list.items()) # 將 cla_dict 寫入 json 文件中json_str = json.dumps(cla_dict, indent=4)with open(’class_indices.json’, ’w’) as json_file: json_file.write(json_str) #訓(xùn)練過程net = AlexNet(num_classes=5, init_weights=True) # 實(shí)例化網(wǎng)絡(luò)（輸出類型為5，初始化權(quán)重）net.to(device) # 分配網(wǎng)絡(luò)到指定的設(shè)備（GPU/CPU）訓(xùn)練loss_function = nn.CrossEntropyLoss() # 交叉熵?fù)p失optimizer = optim.Adam(net.parameters(), lr=0.0002) # 優(yōu)化器（訓(xùn)練參數(shù)，學(xué)習(xí)率） save_path = ’./AlexNet.pth’best_acc = 0.0 for epoch in range(150): ########################################## train ############################################### net.train() # 訓(xùn)練過程中開啟 Dropout running_loss = 0.0# 每個(gè) epoch 都會(huì)對(duì) running_loss 清零 time_start = time.perf_counter()# 對(duì)訓(xùn)練一個(gè) epoch 計(jì)時(shí)for step, data in enumerate(train_loader, start=0): # 遍歷訓(xùn)練集，step從0開始計(jì)算images, labels = data # 獲取訓(xùn)練集的圖像和標(biāo)簽optimizer.zero_grad()# 清除歷史梯度outputs = net(images.to(device)) # 正向傳播loss = loss_function(outputs, labels.to(device)) # 計(jì)算損失loss.backward() # 反向傳播optimizer.step() # 優(yōu)化器更新參數(shù)running_loss += loss.item()# 打印訓(xùn)練進(jìn)度（使訓(xùn)練過程可視化）rate = (step + 1) / len(train_loader) # 當(dāng)前進(jìn)度 = 當(dāng)前step / 訓(xùn)練一輪epoch所需總stepa = '*' * int(rate * 50)b = '.' * int((1 - rate) * 50)with open(os.path.join('train.log'), 'a') as log: log.write(str('rtrain loss: {:^3.0f}%[{}->{}]{:.3f}'.format(int(rate * 100), a, b, loss))+'n')print('rtrain loss: {:^3.0f}%[{}->{}]{:.3f}'.format(int(rate * 100), a, b, loss), end='') print() with open(os.path.join('train.log'), 'a') as log: log.write(str(’%f s’ % (time.perf_counter()-time_start))+'n') print(’%f s’ % (time.perf_counter()-time_start)) ########################################### validate ########################################### net.eval() # 驗(yàn)證過程中關(guān)閉 Dropout acc = 0.0 with torch.no_grad():for val_data in validate_loader: val_images, val_labels = val_data outputs = net(val_images.to(device)) predict_y = torch.max(outputs, dim=1)[1] # 以output中值最大位置對(duì)應(yīng)的索引（標(biāo)簽）作為預(yù)測(cè)輸出 acc += (predict_y == val_labels.to(device)).sum().item() val_accurate = acc / val_num# 保存準(zhǔn)確率最高的那次網(wǎng)絡(luò)參數(shù)if val_accurate > best_acc: best_acc = val_accurate torch.save(net.state_dict(), save_path)with open(os.path.join('train.log'), 'a') as log: log.write(str(’[epoch %d] train_loss: %.3f test_accuracy: %.3f n’ % (epoch + 1, running_loss / step, val_accurate))+'n')print(’[epoch %d] train_loss: %.3f test_accuracy: %.3f n’ % (epoch + 1, running_loss / step, val_accurate))with open(os.path.join('train.log'), 'a') as log: log.write(str(’Finished Training’)+'n')print(’Finished Training’)

訓(xùn)練結(jié)果后，準(zhǔn)確率是94%

訓(xùn)練日志如下：