【王者荣耀】皮肤-英雄 预测（tensorflow）

本项目的思路是通过英雄的多个皮肤进行模型的训练，进而对给定的原皮实施对应英雄的预测。包括的模块有dataset的建立，卷积神经网络模型等。

码云开源地址：aov_ornament-hero: 皮肤进行模型训练原皮进行英雄预测

网盘下载地址：https://pan.quark.cn/s/da3506ed4053

目录

一、思路讲解

二、全部代码

1、main.py

2、dataset.py

3、function2.py

4.create_test_images.py

三、结语

---

一、思路讲解

使用opencv进行图片的RGB矩阵化处理

使用numpy进行多维矩阵的创建与赋值

使用matplotlib进行dataset的预览与测试

使用tensorflow进行神经网络的建立与训练

二、全部代码

1、main.py

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from dataset import datasetss
from tensorflow import keras
from keras import layers, models
from create_test_images import create_test_images

"""提取数据集dataset"""
dataset = datasetss()
# 像素值归一化处理
train_images = dataset.train_images / 255.0
test_images = dataset.test_images / 255.0
train_labels = dataset.train_labels
test_labels = dataset.test_labels


"设置英雄名"
class_names = ['DiRenJie', 'LuBanQiHao',
               'SunShangXiang', 'XiaoQiao', 'DiaoChan']


#"测试数据集"
# plt.figure(figsize=(50, 50))
# plt.imshow(test_images[4])
# plt.show()


"建立模型"
model = models.Sequential()
#卷积层&池化层
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(50, 50, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPool2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
# 展开层
model.add(layers.Flatten())
# 全连接层
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(5))


"编译模型"
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])


"训练模型"
# 拟合数据
model.fit(train_images, train_labels, epochs=10)


"模型预测"
probability_model = tf.keras.Sequential([model, tf.keras.layers.Softmax()])
predictions = probability_model.predict(test_images)

# guess_images =  create_test_images("0.jpg")
#
# probability_model = tf.keras.Sequential([model, tf.keras.layers.Softmax()])
# predictions = probability_model.predict(guess_images)

"结果输出"
print(class_names[np.argmax(predictions[2])])


# "训练模型"
# history = model.fit(train_images, train_labels, epochs=10,
#                     validation_data=(test_images, test_labels))
#
# "评估准确度"
# plt.plot(history.history['accuracy'], label='accuracy')
# plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
# plt.xlabel('Epoch')
# plt.ylabel('Accuracy')
# plt.ylim([0.5, 1])
# plt.legend(loc='lower right')
# plt.show()

2、dataset.py

import functions2


"设置datasetss类"
class datasetss():
    def __init__(self):
        self.train_images = functions2.get_train_images()
        self.train_labels = functions2.get_train_labels()
        self.test_images = functions2.get_test_images()
        self.test_labels = functions2.get_test_labels()

3、function2.py

import cv2 as cv
import numpy as np


"设置train与test标签"
trainfile_labels = ["00", "01", "02", "03", "04", "05",
                    "10", "11", "12", "13", "14", "15", "16", "17",
                    "20", "21", "22", "23", "24", "25", "26",
                    "30", "31", "32", "33", "34", "35",
                    "40", "41", "42", "43", "44"]
testfile_labels = ["0", "1", "2", "3", "4"]


"train_images"
def get_train_images():
    # 构建全零ndarray
    train_images = np.zeros((32, 50, 50, 3))
    i = 0
    for label in trainfile_labels:
        # 选择路径名
        image_path = "D:/WZRY_Project/aov_ornament-hero/images/train_images/" + label + ".jpg"
        # 以RGB模式读取image
        img_ini = cv.imread(image_path, 1)
        # 缩小image到50x50像素
        img_fin = cv.resize(img_ini, (50, 50), interpolation=cv.INTER_CUBIC)
        # 将三维数组赋值给四维数组的后三维
        train_images[i, :, :, :] = img_fin
        i = i + 1
    return train_images


"train_labels"
def get_train_labels():
    # 构建全零ndarray
    train_labels = np.zeros((32, 1))
    i = 0
    for label in trainfile_labels:
        # 将一维数组赋值给二维数组的第二维
        train_labels[i][0] = label[0]
        i = i + 1
    return train_labels


"test_images"
def get_test_images():
    test_images = np.zeros((5, 50, 50, 3))
    i = 0
    for label in testfile_labels:
        image_path = "D:/WZRY_Project/aov_ornament-hero/images/test_images/" + label + ".jpg"
        img_ini = cv.imread(image_path, 1)
        img_fin = cv.resize(img_ini, (50, 50), interpolation=cv.INTER_CUBIC)
        test_images[i, :, :, :] = img_fin
        i = i + 1
    return test_images


"test_labels"
def get_test_labels():
    test_labels = np.zeros((5, 1))
    i = 0
    for label in testfile_labels:
        test_labels[i][0] = label[0]
        i = i + 1
    return test_labels

4.create_test_images.py

import cv2 as cv
import numpy as np


def create_test_images(name):
    test_images = np.zeros((1, 50, 50, 3))
    image_path = "D:/WZRY_Project/aov_ornament-hero/images/guess_images/" + name
    img_ini = cv.imread(image_path, 1)
    img_fin = cv.resize(img_ini, (50, 50), interpolation=cv.INTER_CUBIC)
    test_images[0, :, :, :] = img_fin
    return test_images

三、结语

很遗憾，这个项目的预测并不准确，究其原因便是训练集太小了。然而，其中的各种方法依旧对于新手有借鉴意义。