训练深度学习对数据的预处理

原创已于 2023-03-02 14:17:22 修改 · 211 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#python #cnn #opencv

于 2023-02-28 19:58:54 首次发布

该代码段主要涉及图像处理，特别是为深度学习模型准备数据。它首先定义了一个类别字典，然后遍历指定目录下的子目录（每个子目录代表一类图像），将训练集和测试集的图像路径分别存储在不同的列表中。接着，它计算所有图像的最大尺寸，并将所有图像填充到统一大小，以适应模型输入。最后，处理后的图像被保存到新的目录下，准备用于训练。

import tensorflow as tf
import numpy as np
import os
from random import shuffle
import cv2 as cv

name_dict = {"BF":0,"BK":1,"BL":2,"BR":3,"CF":4,"CL":5,'CV':6,'CXK':7,'S':8,'XF':9}

data_root_path = "color_part_data/"
test_file_path = data_root_path +"test.txt"  #测试集数据集文件
trainer_file_path = data_root_path +"trainer.txt"  #训练集数据集文件

name_data_list = {}  #记录每类图片有多少训练图片、测试图片

trainer_list = []
test_list = []

#将图片完整路径存入字典
def save_train_test_file(path,name):
    if name not in name_data_list:
        img_list =[]
        img_list.append(path)
        name_data_list[name] = img_list
    else:
        name_data_list[name].append(path)

#遍历数据集目录，提取出图片路径，分训练集、测试集
dirs = os.listdir(data_root_path)
for d in dirs:
    full_path = data_root_path + d
    if os.path.isdir(full_path):
        imgs = os.listdir(full_path)  #列出子目录中所有图片
        for img in imgs:
            save_train_test_file(full_path+ "/" + img, d)

#返回图像的所有路径
def AllPicPath(data_root_path):
    IMG = []
    dirs = os.listdir(data_root_path)
    for d in dirs:
        full_path = data_root_path + d
        if os.path.isdir(full_path):
            imgs = os.listdir(full_path)
            for img in imgs:
                IMG.append(full_path+'/'+img)
    return IMG

#计算所有图像最大的长和宽
def MAXLenWid(img):
    leng , wid = 0 , 0
    for imgpath in img:
        image = cv.imread(imgpath)
        L = image.shape[0]
        W = image.shape[1]
        if leng <= L :
           leng = L
        if wid <= W:
           wid = W
    return leng , wid
maxwid , maxleng = MAXLenWid(AllPicPath(data_root_path))
print(maxwid,maxleng,3)

print("---------------------------------------生成最大长宽图像-------------------------------------")
# image_fill = np.zeros((maxleng,maxwid,3))
#定义创建目录函数
def mkdir(path):
    isExists = os.path.exists(path)
    if not isExists:
        os.mkdir(path)
    else:
        print("目录已存在")

allpicpath  = AllPicPath(data_root_path)
for picpath in allpicpath:
    image_fill = np.zeros((maxwid, maxleng, 3))
    print(type(picpath))
    img = cv.imread(picpath)
    print(img.shape)
    wid = img.shape[0]
    leng = img.shape[1]
    offset_w = int((maxwid - wid)/2)
    offset_l = int((maxleng - leng)/2)
    for i in range(3):
        for j in range(wid):
            for k in range(leng):
                image_fill[j + offset_w][k + offset_l][i] = img[j][k][i]
    final_img = image_fill
    picpath1 = picpath.split('.')
    picpath2 = picpath1[0][16:18]
    save_path = 'color_part_data_processing/'+picpath2
    mkdir(save_path)
    cv.imwrite('color_part_data_processing/'+ picpath1[0][16:]+".jpg", final_img, [cv.IMWRITE_PNG_COMPRESSION, 0])