农作物病害识别

##用PlantVillage数据集训练 如下图所示

import os
import cv2
import numpy as np
import pandas as pd
from PIL import Image
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from tensorflow import keras
from keras.models import Sequential
from keras.layers import Conv2D,MaxPooling2D,Dense,Flatten,Dropout
from keras.layers.normalization import BatchNormalization

fpath = "C:/Users/26335/Desktop/DownLoad/Plant/PlantVillage/plantvillage/"
random_seed = 111
categories = os.listdir(fpath)
print(categories)
categories.remove("Pepper__bell___Bacterial_spot")
categories.remove("Potato___healthy")

def load_images_and_labels(categories):
    img_lst=[]
    labels=[]
    for index, category in enumerate(categories):
        for image_name in os.listdir(fpath+"/"+category)[:300]:
            file_ext = image_name.split(".")[-1]
            if (file_ext.lower() == "jpg") or (file_ext.lower() == "jpeg"):
                #print(f"\n Fpath={fpath},Category = {category}, Image name = {image_name}")
                path=fpath+category+"/"+image_name
                #print(path)
                img = cv2.imread(path)
                #print(img)
                img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
                img_array = Image.fromarray(img, 'RGB')

                #resize image to 227 x 227 because the input image resolution for AlexNet is 227 x 227
                resized_img = img_array.resize((227, 227))

                img_lst.append(np.array(resized_img))

                labels.append(index)
    return img_lst, labels

images, labels = load_images_and_labels(categories)
print("No. of images loaded = ",len(images),"\nNo. of labels loaded = ",len(labels))
print(type(images),type(labels))
images = np.array(images)
labels = np.array(labels)
print("Images shape = ",images.shape,"\nLabels shape = ",labels.shape)
print(type(images),type(labels))

def display_rand_images(images, labels):
    plt.figure(1 , figsize = (19 , 10))
    n = 0 
    for i in range(9):
        n += 1 
        r = np.random.randint(0 , images.shape[0] , 1)
        plt.subplot(3 , 3 , n)
        plt.subplots_adjust(hspace = 0.3 , wspace = 0.3)
        plt.imshow(images[r[0]])
        plt.title('Plant label : {}'.format(labels[r[0]]))
        plt.xticks([])
        plt.yticks([])  
    plt.show()
 display_rand_images(images, labels)

#1-step in data shuffling
#get equally spaced numbers in a given range
n = np.arange(images.shape[0])
print("'n' values before shuffling = ",n)
#shuffle all the equally spaced values in list 'n'
np.random.seed(random_seed)
np.random.shuffle(n)
print("\n'n' values after shuffling = ",n)

#2-step in data shuffling
#shuffle images and corresponding labels data in both the lists
images = images[n]
labels = labels[n]
print("Images shape after shuffling = ",images.shape,"\nLabels shape after shuffling = ",labels.shape)

images = images.astype(np.float32)
labels = labels.astype(np.int32)
images = images/255
print("Images shape after normalization = ",images.shape)
display_rand_images(images, labels) 

x_train, x_test, y_train, y_test = train_test_split(images, labels, test_size = 0.2, random_state = random_seed)
print("x_train shape = ",x_train.shape)
print("y_train shape = ",y_train.shape)
print("\nx_test shape = ",x_test.shape)
print("y_test shape = ",y_test.shape)
display_rand_images(x_train, y_train)

model=Sequential()
#1 conv layer
model.add(Conv2D(filters=96,kernel_size=(11,11),strides=(4,4),padding="valid",activation="relu",input_shape=(227,227,3)))

#1 max pool layer
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(BatchNormalization())

#2 conv layer
model.add(Conv2D(filters=256,kernel_size=(5,5),strides=(1,1),padding="valid",activation="relu"))

#2 max pool layer
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(BatchNormalization())

#3 conv layer
model.add(Conv2D(filters=384,kernel_size=(3,3),strides=(1,1),padding="valid",activation="relu"))

#4 conv layer
model.add(Conv2D(filters=384,kernel_size=(3,3),strides=(1,1),padding="valid",activation="relu"))

#5 conv layer
model.add(Conv2D(filters=256,kernel_size=(3,3),strides=(1,1),padding="valid",activation="relu"))

#3 max pool layer
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(BatchNormalization())
model.add(Flatten())

#1 dense layer
model.add(Dense(4096,input_shape=(227,227,3),activation="relu"))
model.add(Dropout(0.4))
model.add(BatchNormalization())

#2 dense layer
model.add(Dense(4096,activation="relu"))
model.add(Dropout(0.4))
model.add(BatchNormalization())

#3 dense layer
model.add(Dense(1000,activation="relu"))
model.add(Dropout(0.4))
model.add(BatchNormalization())

#output layer
model.add(Dense(20,activation="softmax"))
model.summary()
model.compile(optimizer="adam", loss="sparse_categorical_crossentropy", metrics=["accuracy"])
history=model.fit(x_train, y_train, epochs=100)

pred = model.predict(x_test)
pred.shape
plt.figure(1 , figsize = (19 , 10))
n = 0 
for i in range(9):
    n += 1 
    r = np.random.randint( 0, x_test.shape[0], 1)
    
    plt.subplot(3, 3, n)
    plt.subplots_adjust(hspace = 0.3, wspace = 0.3)
    
    plt.imshow(x_test[r[0]])
    
    plt.title('Actual = {}, Predicted = {}'.format(y_test[r[0]] , pred[r[0]][y_test[r[0]]]) )
    plt.xticks([]) , plt.yticks([])
    #print('Plant label : {}'.format(labels[r[0]]))
    #print(pred[r[0]])
    print('r[0]={},y_test[r[0]]={},labels[r[0]]={}'.format(r[0],y_test[r[0]],labels[r[0]]))
plt.show()

print("Saving the model......")
# save model in JSON format
model_json = model.to_json()
json_file = open("C:/Users/26335/Desktop/DownLoad/Plant/working/model1.json", "w")
json_file.write(model_json)
print("Model saved in JSON format!")
# save training weights in h5 file
model.save_weights("C:/Users/26335/Desktop/DownLoad/Plant/working/model1.h5")
print("\nModel weights saved!")
%cd C:/Users/26335/Desktop/DownLoad/Plant/working
from IPython.display import FileLink
FileLink(r'model1.h5')

acc = history.history['accuracy']
loss = history.history['loss']
epochs = range(1, len(acc) + 1)
#Train and validation accuracy
plt.plot(epochs, acc, 'b', label='Training accurarcy')
plt.plot(epochs, loss, 'r', label='Training loss')
plt.title('Training and Validation accurarcy')
plt.legend()
plt.show()