我要评论
前言：
最近做了一个农作物虫害图像识别的程序，在此分享一下。本文用到的深度学习框架为tensorflow2，opencv等等！使用的数据集共有61种类别，分别代表不同的虫害类别。使用的网络模型为moblienetv3.
bi 设 dai 坐
效果视频如下所示：
农作物虫害图像识别
搭建mobilenetv3模型
代码如下所示：
# 根据tf.keras的官方代码修改的mobilenetv3的网络模型
import tensorflow as tf
from keras import layers, models
"""
reference:
- [searching for mobilenetv3](https://arxiv.org/pdf/1905.02244.pdf) (iccv 2019)
the following table describes the performance of mobilenets v3:
------------------------------------------------------------------------
macs stands for multiply adds
|classification checkpoint|macs(m)|parameters(m)|top1 accuracy|pixel1 cpu(ms)|
|---|---|---|---|---|
| mobilenet_v3_large_1.0_224 | 217 | 5.4 | 75.6 | 51.2 |
| mobilenet_v3_large_0.75_224 | 155 | 4.0 | 73.3 | 39.8 |
| mobilenet_v3_large_minimalistic_1.0_224 | 209 | 3.9 | 72.3 | 44.1 |
| mobilenet_v3_small_1.0_224 | 66 | 2.9 | 68.1 | 15.8 |
| mobilenet_v3_small_0.75_224 | 44 | 2.4 | 65.4 | 12.8 |
| mobilenet_v3_small_minimalistic_1.0_224 | 65 | 2.0 | 61.9 | 12.2 |
for image classification use cases, see
[this page for detailed examples](https://keras.io/api/applications/#usage-examples-for-image-classification-models).
for transfer learning use cases, make sure to read the
[guide to transfer learning & fine-tuning](https://keras.io/guides/transfer_learning/).
"""
##################################################################################################################################
# 定义v3的完整模型 #################################################################################################################
##################################################################################################################################
def mobilenetv3(input_shape=[224, 224, 3], classes=1000, dropout_rate=0.2, alpha=1.0, weights=none,
model_type='large', minimalistic=false, classifier_activation='softmax', include_preprocessing=false):
# 如果有权重文件，那就意味着要迁移学习，那就意味着需要让bn层始终处于infer状态，否则解冻整个网络后，会出现acc下降loss上升的现象，终其原因是解冻网络之
# 前，网络bn层用的是之前数据集的均值和方差，解冻后虽然维护着新的滑动平均和滑动方差，但是单次训练时使用的是当前batch的均值和方差，差异太大造成特征崩塌
if weights:
bn_training = false
else:
bn_training = none
bn_decay = 0.99 # bn层的滑动平均系数，这个值的设置需要匹配steps和batchsize否则会出现奇怪现象
# 确定通道所处维度
channel_axis = -1
# 根据是否为mini设置，修改部分配置参数
if minimalistic:
kernel = 3
activation = relu
se_ratio = none
name = "mini"
else:
kernel = 5
activation = hard_swish
se_ratio = 0.25
name = "norm"
# 定义模型输入张量
img_input = layers.input(shape=input_shape)
# 是否包含预处理层
if include_preprocessing:
x = layers.rescaling(scale=1. / 127.5, offset=-1.)(img_input)
else:
x = img_input
# 定义整个模型的第一个特征提取层
x = layers.conv2d(16, kernel_size=3, strides=(2, 2), padding='same', use_bias=false, name='conv')(x)
x = layers.batchnormalization(axis=channel_axis, epsilon=1e-3, momentum=bn_decay, name='conv/batchnorm')(x,
training=bn_training)
x = activation(x)
# 定义整个模型的骨干特征提取
if model_type == 'large':
x = mobilenetv3large(x, kernel, activation, se_ratio, alpha, bn_training, bn_decay)
last_point_ch = 1280
else:
x = mobilenetv3small(x, kernel, activation, se_ratio, alpha, bn_training, bn_decay)
last_point_ch = 1024
# 定义整个模型的后特征提取
last_conv_ch = _depth(x.shape[channel_axis] * 6)
# if the width multiplier is greater than 1 we increase the number of output channels
if alpha > 1.0:
last_point_ch = _depth(last_point_ch * alpha)
x = layers.conv2d(last_conv_ch, kernel_size=1, padding='same', use_bias=false, name='conv_1')(x)
x = layers.batchnormalization(axis=channel_axis, epsilon=1e-3, momentum=bn_decay, name='conv_1/batchnorm')(x,
training=bn_training)
x = activation(x)
# 如果tf版本大于等于2.6则直接使用下面第一句就可以了，否则使用下面2~3句
# x = layers.globalaveragepooling2d(data_format='channels_last', keepdims=true)(x)
x = layers.globalaveragepooling2d(data_format='channels_last')(x)
x = tf.expand_dims(tf.expand_dims(x, 1), 1)
# 定义第一个特征分类层
x = layers.conv2d(last_point_ch, kernel_size=1, padding=
发表评论