遗传算法优化卷积神经网络（人脸识别分类）.zip

# 主函数 #导入要用到的模块 import paddle import paddle.fluid as fluid import numpy import sys from multiprocessing import cpu_count import matplotlib.pyplot as plt import random import operator import os import json # ①testrate取0.05,0.1,0.2 def setDataList(test_rate): data_root_path = '/home/aistudio/data/data2394/images/face' class_detail = [] class_dirs = os.listdir(data_root_path) class_label = 0 father_paths = data_root_path.split('/') while True: if father_paths[father_paths.__len__() - 1] == '': del father_paths[father_paths.__len__() - 1] else: break father_path = father_paths[father_paths.__len__() - 1] data_list_path = '/home/aistudio/data/data2394/%s/' % father_path isexist = os.path.exists(data_list_path) if not isexist: os.makedirs(data_list_path) with open(data_list_path + "test.list", 'w') as f: # f.truncate() pass with open(data_list_path + "trainer.list", 'w') as f: # f.truncate() pass # 总的图像数量 all_class_images = 0 # 读取每个类别 for class_dir in class_dirs:#["jiangwen","pengyuyan","zhangziyi"] # 每个类别的信息 class_detail_list = {} test_sum = 0 trainer_sum = 0 # 统计每个类别有多少张图片 class_sum = 0 # 获取类别路径 path = data_root_path + "/" + class_dir # 获取所有图片 img_paths = os.listdir(path) for img_path in img_paths: # 遍历文件夹下的每个图片 name_path = path + '/' + img_path # 每张图片的路径 if class_sum % (1.0/test_rate) == 0: # 每10张图片取一个做测试数据 test_sum += 1 #test_sum测试数据的数目 with open(data_list_path + "test.list", 'a') as f:#追加内容到test.list f.write(name_path + "\t%d" % class_label + "\n") #class_label 标签：0,1,2 else: trainer_sum += 1 #trainer_sum测试数据的数目 with open(data_list_path + "trainer.list", 'a') as f: f.write(name_path + "\t%d" % class_label + "\n")#class_label 标签：0,1,2 class_sum += 1 #每类图片的数目 all_class_images += 1 #所有类图片的数目 # 说明的json文件的class_detail数据 class_detail_list['class_name'] = class_dir #类别名称，如jiangwen class_detail_list['class_label'] = class_label #类别标签，0,1,2 class_detail_list['class_test_images'] = test_sum #该类数据的测试集数目 class_detail_list['class_trainer_images'] = trainer_sum #该类数据的训练集数目 class_detail.append(class_detail_list) class_label += 1 #class_label 标签：0,1,2 # 获取类别数量 all_class_sum = class_dirs.__len__() # 说明的json文件信息 readjson = {} readjson['all_class_name'] = father_path #文件父目录 readjson['all_class_sum'] = all_class_sum # readjson['all_class_images'] = all_class_images readjson['class_detail'] = class_detail jsons = json.dumps(readjson, sort_keys=True, indent=4, separators=(',', ': ')) with open(data_list_path + "readme.json",'w') as f: f.write(jsons) print ('生成数据列表完成！') # 定义训练的mapper # train_mapper函数的作用是用来对训练集的图像进行处理修剪和数组变换，返回img数组和标签 # sample是一个python元组，里面保存着图片的地址和标签。 ('../images/face/zhangziyi/20181206145348.png', 2) def train_mapper(sample): img, label = sample # 进行图片的读取，由于数据集的像素维度各不相同，需要进一步处理对图像进行变换 img = paddle.dataset.image.load_image(img) #进行了简单的图像变换，这里对图像进行crop修剪操作，输出img的维度为(3, 100, 100) img = paddle.dataset.image.simple_transform(im=img, #输入图片是HWC resize_size=100, # 剪裁图片 crop_size=100, is_color=True, #彩色图像 is_train=True) #将img数组进行进行归一化处理，得到0到1之间的数值 img= img.flatten().astype('float32')/255.0 return img, label # 对自定义数据集创建训练集train的reader def train_r(train_list, buffered_size=1024): def reader(): with open(train_list, 'r') as f: # 将train.list里面的标签和图片的地址方法一个list列表里面，中间用\t隔开' #../images/face/jiangwen/0b1937e2-f929-11e8-8a8a-005056c00008.jpg\t0' lines = [line.strip() for line in f] for line in lines: # 图像的路径和标签是以\t来分割的,所以我们在生成这个列表的时候,使用\t就可以了 img_path, lab = line.strip().split('\t') yield img_path, int(lab) # 创建自定义数据训练集的train_reader return paddle.reader.xmap_readers(train_mapper, reader,cpu_count(), buffered_size) # sample是一个python元组，里面保存着图片的地址和标签。 ('../images/face/zhangziyi/20181206145348.png', 2) def test_mapper(sample): img, label = sample img = paddle.dataset.image.load_image(img) img = paddle.dataset.image.simple_transform(im=img, resize_size=100, crop_size=100, is_color=True, is_train=False) img= img.flatten().astype('float32')/255.0 return img, label # 对自定义数据集创建验证集test的reader def test_r(test_list, buffered_size=1024): def reader(): with open(test_list, 'r') as f: lines = [line.strip() for line in f] for line in lines: #图像的路径和标签是以\t来分割的,所以我们在生成这个列表的时候,使用\t就可以了 img_path, lab = line.strip().split('\t') yield img_path, int(lab) return paddle.reader.xmap_readers(test_mapper, reader,cpu_count(), buffered_size) # 打印测试 # temp_reader = paddle.batch(trainer_reader, # batch_size=3) # temp_data=next(temp_reader()) # print(temp_data) def convolutional_neural_network(image, type_size): # 第一个卷积--池化层 conv_pool_1 = fluid.nets.simple_img_conv_pool(input=image,# 输入图像 filter_size=3,# 滤波器的大小 num_filters=32,# filter 的数量。它与输出的通道相同 pool_size=2,# 池化层大小2*2 pool_stride=2,# 池化层步长 act='relu') # 激活类型 # Dropout主要作用是减少过拟合，随机让某些权重不更新 # Dropout是一种正则化技术，通过在训练过程中阻止神经元节点间的联合适应性来减少过拟合。 # 根据给定的丢弃概率dropout随机将一些神经元输出设置为0，其他的仍保持不变。 drop = fluid.layers.dropout(x=conv_pool_1, dropout_prob=0.5) # 第二个卷积--池化层 conv_pool_2 = fluid.nets.simple_img_conv_pool(input=drop, filter_size=3,