飞桨告诉我谁是最勤劳的小蜜蜂

谁是最勤劳的小蜜蜂

蜜蜂生来就是为了采蜜，它也是很辛勤，不是在采蜜就是已经采完蜜了，要不然是在蜂巢里吐蜜，直到奉献自己的一生。据悉勤劳的密封翅膀都带花粉的，那我们就来瞅瞅吧。

反正我肉眼看不清他们翅膀、小腿上的花粉，那么就交给机器来解决吧，机器学习嘛。

import os
import zipfile
import random
import json
import paddle
import sys
import numpy as np
from PIL import Image
from PIL import ImageEnhance
import paddle
import matplotlib.pyplot as plt
%matplotlib inline 

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/__init__.py:107: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import MutableMapping
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/rcsetup.py:20: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import Iterable, Mapping
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/colors.py:53: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import Sized

一、数据准备

• （1）解压原始数据集
• （2）按照比例划分训练集与验证集
• （3）乱序，生成数据列表
• （4）构造训练数据集提供器和验证数据集提供器

# 已解压可注释
# !unzip -q data/data71008/花粉数据集archive.zip -d dataset

import paddle
import paddle.vision.transforms as T
import numpy as np
from PIL import Image


class BeeDataset(paddle.io.Dataset):
    """
    2类Bee数据集类的定义
    """
    def __init__(self,mode='train',rate=0.2):
        """
        初始化函数
        """
        self.all_data = []
        self.data = []

        with open('dataset/PollenDataset/pollen_data.csv') as f:
            next(f)
            for line in f.readlines():
                info = line.strip().split(',')
                if len(info) > 0:
                    self.all_data.append([info[1].strip(), info[2].strip()])
        self.transforms = T.Compose([
            T.Resize((64,64)),    # （h，w） 180,300 图片缩放
            T.ToTensor(),                       # 数据的格式转换和标准化、 HWC => CHW    
            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])                    
            ])
        if mode=='train':
            self.data=self.all_data[int(len(self.all_data)*rate):len(self.all_data)]
        else:
            self.data=self.all_data[0:int(len(self.all_data)*rate)]

    def get_origin_data(self):
        return self.data

    def __getitem__(self, index):
        """
        根据索引获取单个样本
        """
        image_file, label = self.data[index]
        image_file=os.path.join('dataset/PollenDataset/images',image_file)
        image = Image.open(image_file)
        if image.mode != 'RGB':
            image = image.convert('RGB')
        image = self.transforms(image)
        return image, np.array(label, dtype='int64')

    def __len__(self):
        """
        获取样本总数
        """
        return len(self.data)

bee=BeeDataset()
train_dataset=BeeDataset(mode='train',rate=0.3)
test_dataset=BeeDataset(mode='test',rate=0.3)
print('train_data len: {}, test_data len:{}'.format(train_dataset.__len__(), test_dataset.__len__()))

train_data len: 500, test_data len:214

二、模型配置

复习下卷积嘛，用啥摩托车，手写好了。

#定义卷积网络
import paddle.nn as nn
import paddle.nn.functional as F
from visualdl import LogWriter

class MyCNN(nn.Layer):
    def __init__(self):
        super(MyCNN,self).__init__()
        self.hidden1 = nn.Conv2D(in_channels=3,       #通道数
                                            out_channels=64,       #卷积核个数
                                            kernel_size =3,        #卷积核大小
                                            stride=1)             #步长
        self.hidden2 = nn.Conv2D(in_channels=64,
                                            out_channels = 128,
                                            kernel_size =3,
                                            stride=1)
        self.hidden3 = nn.MaxPool2D(kernel_size=2,          #池化核大小
                                            stride=2)       #池化步长
        self.hidden4 = nn.Linear(in_features=128*30*30,out_features=2)
    #网络的前向计算过程
    def forward(self,input):
        x = self.hidden1(input)
        with LogWriter(logdir="./chk_points/conv1/") as writer:
            for i in range(10):
                myimg=x[i][0:3].numpy()
                print(myimg.shape)
                print(myimg)
                writer.add_image(tag='conv1',img=myimg, step=i)
        x=F.relu(x)
        # print(x.shape)
        x = self.hidden2(x)
        x=F.relu(x)
        # print(x.shape)
        x = self.hidden3(x)
        # print(x.shape)
        #卷积层的输出特征图如何当作全连接层的输入使用呢？
        #卷积层的输出数据格式是[N,C,H,W],在输入全连接层的时候，会自动将数据拉平.
        #也就是对每个样本，自动将其转化为长度为K的向量，其中K=C×H×W，一个mini-batch的数据维度变成了N×K的二维向量。
        x = paddle.reshape(x, shape=[-1, 128*30*30])
        x = self.hidden4(x)
        out = F.softmax(x)        
        return out

三、模型训练 && 四、模型评估

import paddle
from paddle import Model
myCNN=MyCNN()
model= Model(myCNN)
model.summary((1,3, 64, 64))

(3, 62, 62)
[[[ 0.7852405   1.1204937   1.0803782  ...  0.79842985  0.6884388
    0.8554362 ]
  [ 0.67254984  1.4634426   1.4180253  ...  1.1190094   0.7330806
    0.36009166]
  [ 0.77952844  1.25177     0.62106866 ...  0.781536    0.5499747
    1.185901  ]
  ...
  [ 1.1787175   1.165362    0.4543565  ...  0.45226488  1.7040001
    1.3579762 ]
  [ 1.3305937   0.9702562   0.6092412  ...  0.56712335  1.6329365
    0.7801384 ]
  [ 0.5835558   0.7375888   1.1013889  ...  1.4835536   0.8401654
    2.0126145 ]]

 [[-1.8209429  -1.010055   -2.013126   ... -0.6853134  -1.2335591
   -1.0887115 ]
  [-1.1185786  -1.1789582  -1.3160777  ... -1.0062238  -1.5497901
   -0.7702041 ]
  [-1.0398924  -1.2710589  -0.8835699  ... -1.412867   -0.75782824
   -0.66315925]
  ...
  [-0.45574644 -0.63937426 -0.54133946 ... -0.7855961  -0.99309015
   -0.50211173]
  [-1.2958944  -1.4259287  -1.220675   ... -0.6283496  -0.8967111
   -0.879598  ]
  [-0.4558576  -0.9036231  -0.61986226 ... -1.2891432  -0.8856837
   -1.3334211 ]]

 [[-0.16028228 -1.4461167  -0.8729783  ...  0.03628984  0.48390952
   -0.23797147]
  [-0.1558084  -0.35822743 -0.15980428 ... -0.86909014  0.12811811
   -0.19449231]
  [-1.0637001  -0.90246856 -0.6251344  ...  0.20142564 -0.9001509
   -0.6823548 ]
  ...
  [-0.49504218  0.37776464 -0.37616605 ... -0.1840626  -0.74749
   -0.11741641]
  [-0.7924826  -0.7401607  -0.12474468 ... -0.9140016  -0.7107928
   -0.23936886]
  [-1.1824067  -0.1575087  -0.43380502 ... -0.540684   -0.65324664
   -0.5257259 ]]]



---------------------------------------------------------------------------

error                                     Traceback (most recent call last)

<ipython-input-40-198a0f8bb235> in <module>
      3 myCNN=MyCNN()
      4 model= Model(myCNN)
----> 5 model.summary((1,3, 64, 64))


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/hapi/model.py in summary(self, input_size, dtype)
   1879         else:
   1880             _input_size = self._inputs
-> 1881         return summary(self.network, _input_size, dtype)
   1882 
   1883     def _verify_spec(self, specs, shapes=None, dtypes=None, is_input=False):


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/hapi/model_summary.py in summary(net, input_size, dtypes)
    147 
    148     _input_size = _check_input(_input_size)
--> 149     result, params_info = summary_string(net, _input_size, dtypes)
    150     print(result)
    151 


</opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/decorator.py:decorator-gen-342> in summary_string(model, input_size, dtypes)


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/dygraph/base.py in _decorate_function(func, *args, **kwargs)
    313         def _decorate_function(func, *args, **kwargs):
    314             with self:
--> 315                 return func(*args, **kwargs)
    316 
    317         @decorator.decorator


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/hapi/model_summary.py in summary_string(model, input_size, dtypes)
    274 
    275     # make a forward pass
--> 276     model(*x)
    277 
    278     # remove these hooks


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/dygraph/layers.py in __call__(self, *inputs, **kwargs)
    889                 self._built = True
    890 
--> 891             outputs = self.forward(*inputs, **kwargs)
    892 
    893             for forward_post_hook in self._forward_post_hooks.values():


<ipython-input-39-e30af9085378> in forward(self, input)
     26                 print(myimg.shape)
     27                 print(myimg)
---> 28                 writer.add_image(tag='conv1',img=myimg, step=i)
     29         x=F.relu(x)
     30         # print(x.shape)


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/visualdl/writer/writer.py in add_image(self, tag, img, step, walltime, dataformats)
    191         self._get_file_writer().add_record(
    192             image(tag=tag, image_array=img, step=step, walltime=walltime,
--> 193                   dataformats=dataformats))
    194 
    195     def add_text(self, tag, text_string, step=None, walltime=None):


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/visualdl/component/base_component.py in image(tag, image_array, step, walltime, dataformats)
    169     image_array = denormalization(image_array)
    170     image_array = convert_to_HWC(image_array, dataformats)
--> 171     image_bytes = imgarray2bytes(image_array)
    172     image = Record.Image(encoded_image_string=image_bytes)
    173     return Record(values=[


/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/visualdl/component/base_component.py in imgarray2bytes(np_array)
     70         import cv2
     71 
---> 72         np_array = cv2.cvtColor(np_array, cv2.COLOR_BGR2RGB)
     73         ret, buf = cv2.imencode(".png", np_array)
     74         img_bin = Image.fromarray(np.uint8(buf)).tobytes("raw")


error: OpenCV(4.1.1) /io/opencv/modules/imgproc/src/color.simd_helpers.hpp:92: error: (-2:Unspecified error) in function 'cv::impl::{anonymous}::CvtHelper<VScn, VDcn, VDepth, sizePolicy>::CvtHelper(cv::InputArray, cv::OutputArray, int) [with VScn = cv::impl::{anonymous}::Set<3, 4>; VDcn = cv::impl::{anonymous}::Set<3, 4>; VDepth = cv::impl::{anonymous}::Set<0, 2, 5>; cv::impl::{anonymous}::SizePolicy sizePolicy = (cv::impl::<unnamed>::SizePolicy)2u; cv::InputArray = const cv::_InputArray&; cv::OutputArray = const cv::_OutputArray&]'
> Invalid number of channels in input image:
>     'VScn::contains(scn)'
> where
>     'scn' is 62

# 模型训练配置
model.prepare(optimizer=paddle.optimizer.Adam(learning_rate=0.00001,parameters=model.parameters()),# 优化器
              loss=paddle.nn.CrossEntropyLoss(),           # 损失函数
              metrics=paddle.metric.Accuracy()) # 评估指标
# 训练可视化VisualDL工具的回调函数
visualdl = paddle.callbacks.VisualDL(log_dir='visualdl_log')              

# 启动模型全流程训练
model.fit(train_dataset,            # 训练数据集
          # test_dataset,            # 评估数据集
          epochs=100,            # 总的训练轮次
          batch_size=256,    # 批次计算的样本量大小
          shuffle=True,             # 是否打乱样本集
          verbose=1,                # 日志展示格式
          save_dir='./chk_points/', # 分阶段的训练模型存储路径
          callbacks=[visualdl])     # 回调函数使用

###模型存储

将我们训练得到的模型进行保存，以便后续评估和测试使用。

model.evaluate(eval_data=test_dataset, verbose=2)

model.save('model_save_dir')

五、模型预测

print('测试数据集样本量：{}'.format(len(test_dataset)))

# 执行预测
result = model.predict(test_dataset)

# 样本映射
LABEL_MAP = ['偷懒的小蜜蜂','勤劳的小蜜蜂']

# 随机取样本展示
indexs = [2, 38, 56, 92, 100, 101]

for idx in range(test_dataset.__len__()):
    predict_label = np.argmax(result[0][idx])
    real_label = test_dataset.__getitem__(idx)[1]
    print('样本ID：{}, 真实标签：{}, 预测值：{}'.format(idx, LABEL_MAP[real_label], LABEL_MAP[predict_label]))

origin_data=test_dataset.get_origin_data()
print(origin_data.__len__())

# 定义画图方法
from PIL import Image
import matplotlib.font_manager as font_manager

fontpath = 'SIMHEI.TTF'
font = font_manager.FontProperties(fname=fontpath, size=10)

def show_img(img, predict):
    plt.figure()
    plt.title(predict, FontProperties=font)
    plt.imshow(img, cmap=plt.cm.binary)
    plt.show()

# 抽样展示
for i in range(10):
    img_path='dataset/PollenDataset/images/' + origin_data[i][0]
    real_label=int(origin_data[i][1])
    predict_label= int(np.argmax(result[0][i]))
    img=Image.open(img_path)
    title='样本ID：{}, 真实标签：{}, 预测值：{}'.format(idx, LABEL_MAP[real_label], LABEL_MAP[predict_label])
    show_img(img, title)
format(idx, LABEL_MAP[real_label], LABEL_MAP[predict_label])
    show_img(img, title)

奥奥，看起来小蜜蜂们都很勤劳啊，哈哈哈。

https://aistudio.baidu.com/aistudio/projectdetail/1549057