淘先锋技术网

首页 1 2 3 4 5 6 7

CNN花卉识别模型

照片链接：6种花卉 每种200+张http://xn--https-bl8js66z7n7i//pan.baidu.com/s/1fVXFHKQqiSL6zuEgx2T27w%20%20%E6%8F%90%E5%8F%96%E7%A0%81%EF%BC%9Aa56t若是先要跟多照片可以使用下方链接去爬取照片

不过爬取完的照片存在一些无用照片需要清洗一下

爬虫https://blog.csdn.net/weixin_52486467/article/details/127152598CNN模型简单介绍一下：

使用的是AlexNet模型：一共五层网络层和3个全连接层

Model(
  (layer1): Sequential(
    (0): Conv2d(3, 96, kernel_size=(8, 8), stride=(4, 4))
    (1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (2): ReLU(inplace=True)
  )
  (layer2): Sequential(
    (0): Conv2d(96, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (2): ReLU(inplace=True)
  )
  (layer3): Sequential(
    (0): Conv2d(256, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
  )
  (layer4): Sequential(
    (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
  )
  (layer5): Sequential(
    (0): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (fc1): Sequential(
    (0): Flatten(start_dim=1, end_dim=-1)
    (1): Linear(in_features=12544, out_features=4096, bias=True)
  )
  (fc2): Linear(in_features=4096, out_features=4096, bias=True)
  (fc3): Linear(in_features=4096, out_features=6, bias=True)
)

Cnov2d:2D卷积 ，主要参数在上方都使用了 

Maxpool2d: 池化，

作用的查看下方连接有介绍到：

CNN神经网络https://blog.csdn.net/weixin_52486467/article/details/126141503

代码：

import numpy as np 
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms,datasets
from torch.utils.data import DataLoader,dataset
import os
import cv2
from PIL import Image
import matplotlib.pyplot as plt
#数据准备
EPOCHS = 10
BATCH_SIZE = 16
LR = 0.001
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu' #能使用GPU便使用GPU  
transform = transforms.Compose([
    transforms.Resize((244,244)),
    transforms.ToTensor(),
    transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))
])
train_ds = datasets.ImageFolder('../Flower_picture/train-dataset/',transform)
test_ds = datasets.ImageFolder('../Flower_picture/test-dataset/',transform)
train_dl = DataLoader(train_ds,BATCH_SIZE,shuffle=True)
test_dl = DataLoader(test_ds,BATCH_SIZE,shuffle=False)
#定义模型
class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.Sequential(
            nn.Conv2d(3,96,kernel_size=8,stride=4,padding=0),#输出（96，55，55）
            nn.MaxPool2d(2),#输出（96，27，27）
            nn.ReLU(inplace=True),
        )
        self.layer2 = nn.Sequential(
            nn.Conv2d(96,256,5,1,2), #输出（256，27，27）
            nn.MaxPool2d(2),         #输出（256，13，13）
            nn.ReLU(inplace=True),
        )
        self.layer3 = nn.Sequential(
            nn.Conv2d(256,384,3,1,1), #输出（384，13，13）
            nn.ReLU(inplace=True)
        )
        self.layer4 = nn.Sequential(
            nn.Conv2d(384,384,3,1,1), #输出（384，13，13）
            nn.ReLU(inplace=True)
        )
        self.layer5 = nn.Sequential(
            nn.Conv2d(384,256,3,1,1),#输出（256，13，13）
            nn.MaxPool2d(2),   #输出（256，6，6）
        )
        #全连接层
        self.fc1 = nn.Sequential(
            nn.Flatten(),
            nn.Linear(256*7*7,4096))
        self.fc2 = nn.Linear(4096,4096)
        self.fc3 = nn.Linear(4096,6)
        
    def forward(self,x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.layer5(x)

#         x = x.reshape(-1,256*7*7)
        x = self.fc1(x)
        x = self.fc2(x)
        x = self.fc3(x)
        
        return x
#测试模型是否能运行  单独运行后注释掉
net = Model()
net.to(DEVICE)
loss_fn = nn.CrossEntropyLoss()
opt = torch.optim.SGD(net.parameters(),lr=LR,momentum=0.9)
imgs, lables = next(iter(train_dl))
net(imgs.to(DEVICE))
#定义训练函数
def fif(epoch,model,device,train_dl,test_dl,loss_fn,opt):
    loss_,acc,correct = 0,0,0
    size = len(train_ds)
    total = len(test_ds)
    for idx,data in enumerate(train_dl):
        imgs, lables = data
        imgs,lables = imgs.to(device),lables.to(device)#加载到内存
        pred = model(imgs)

        loss = loss_fn(pred,lables)
#         print('loss:',loss)
#         print('label:',lables)
        opt.zero_grad()
        loss.backward()
        opt.step()
        loss_+=loss
#         print('idx:',idx)
        if idx%10==9:
            print(f'epoch:{epoch+1} {idx+1} loss:{loss_/10:.3f}')
            loss =0 #每次清零
        acc +=(pred.argmax(1)==lables).type(torch.float).sum().item()
    print(f'【当前的准确率：{acc/size:.3f}】')
    
    
    with torch.no_grad():
        for idx,data in enumerate(test_dl):
            imgs, lables = data
            imgs,lables = imgs.to(device),lables.to(device)
            pred = model(imgs)
            correct+=(pred.argmax(1)==lables).type(torch.float).sum().item()
        print(f'【test当前准确率：{correct/total:.3f}】')
#调用训练函数开始训练
for epoch in range(EPOCHS):
    fif(epoch,net,DEVICE,train_dl,test_dl,loss_fn,opt)
#定义字典 以便下面输出预测值
flower_dic = dict(zip(test_ds.class_to_idx.values(), test_ds.class_to_idx.keys()))
img_path = '../Flower_picture/test-dataset/dandelion/00b56fea1ec68d8e6019a167073a623d.jpeg' #自行更改路径
img  = Image.open(img_path).convert('RGB')
plt.imshow(img)
img = transform(img)
img = torch.reshape(img, (1, 3, 244, 244))
flower_dic[net(img.to(DEVICE)).argmax(1).item()]

模型还可以优化：

1、在每个网络层添加激活函数

2、test_ds 测试集我使用的照片只有两张  正常情况测试集取全部照片的0.2-0.3之间