目录
1. 打开Anaconda Prompt（Anaconda3）
1. 项目背景
瓷绝缘子在长期机电负荷与恶劣气候条件影响下，易发生劣化并出现零值现象，导致绝缘子串的有效爬电距离缩短，在过电压下易发生闪络击穿，严重威胁输配电线路的安全运行。红外热像法因其非接触式、安全高效的优点，成为现有技术中相对可行的零值绝缘子带电检测方法。
2. 图像数据集介绍
收集到440幅绝缘子红外测温图像，并利用labelimg对其中的绝缘子进行标注。
绝缘子VOC标签的格式如下：
-
<annotation>
-
<folder></folder>
-
<filename>
1.jpg</filename>
-
<path>
1.jpg</path>
-
<source>
-
<database>roboflow.ai</database>
-
</source>
-
<size>
-
<width>
416</width>
-
<height>
416</height>
-
<depth>
3</depth>
-
</size>
-
<segmented>
0</segmented>
-
<
object>
-
<name>insulator</name>
-
<pose>Unspecified</pose>
-
<truncated>
0</truncated>
-
<difficult>
0</difficult>
-
<occluded>
0</occluded>
-
<bndbox>
-
<xmin>
231</xmin>
-
<xmax>
405</xmax>
-
<ymin>
1</ymin>
-
<ymax>
212</ymax>
-
</bndbox>
-
</
object>
-
<
object>
-
<name>insulator</name>
-
<pose>Unspecified</pose>
-
<truncated>
0</truncated>
-
<difficult>
0</difficult>
-
<occluded>
0</occluded>
-
<bndbox>
-
<xmin>
193</xmin>
-
<xmax>
254</xmax>
-
<ymin>
262</ymin>
-
<ymax>
367</ymax>
-
</bndbox>
-
</
object>
-
</annotation>
labelimg的安装流程：
1. 打开Anaconda Prompt（Anaconda3）
2. 创建一个新环境来安装labelimg
conda create -n labelimg python=3.73. 激活新创建的环境labelimg
conda activate labelimg4.输入
conda activate labelimg5.输入labelimg 即可运行
3. 模型介绍
模型训练过程中采用余弦退火衰减算法、并采用adam优化完成对模型权值参数的更新，其中训练集：测试集=9:1，冻结原始预训练YOLOv5模型前234层，迭代训练800个epoch，batchsize1=8，再解冻迭代训练200个epoch，batchsize2=4，3060ti显卡。
主干特征提取网络：
-
from functools
import wraps
-
-
import tensorflow
as tf
-
from tensorflow.keras
import backend
as K
-
from tensorflow.keras.initializers
import RandomNormal
-
from tensorflow.keras.layers
import (Add, BatchNormalization, Concatenate,
-
Conv2D, Layer, MaxPooling2D,
-
ZeroPadding2D)
-
from tensorflow.keras.regularizers
import l2
-
from utils.utils
import compose
-
-
-
class
SiLU(
Layer):
-
def
__init__(
self, **kwargs):
-
super(SiLU, self).__init__(**kwargs)
-
self.supports_masking =
True
-
-
def
call(
self, inputs):
-
return inputs * K.sigmoid(inputs)
-
-
def
get_config(
self):
-
config =
super(SiLU, self).get_config()
-
return config
-
-
def
compute_output_shape(
self, input_shape):
-
return input_shape
-
-
class
Focus(
Layer):
-
def
__init__(
self):
-
super(Focus, self).__init__()
-
-
def
compute_output_shape(
self, input_shape):
-
return (input_shape[
0], input_shape[
1] //
2
if input_shape[
1] !=
None
else input_shape[
1], input_shape[
2] //
2
if input_shape[
2] !=
None
else input_shape[
2], input_shape[
3] *
4)
-
-
def
call(
self, x):
-
return tf.concat(
-
[x[..., ::
2, ::
2, :],
-
x[...,
1::
2, ::
2, :],
-
x[..., ::
2,
1::
2, :],
-
x[...,
1::
2,
1::
2, :]],
-
axis=-
1
-
)
-
-
@wraps(Conv2D)
-
def
DarknetConv2D(
*args, **kwargs):
-
darknet_conv_kwargs = {
'kernel_initializer' : RandomNormal(stddev=
0.02),
'kernel_regularizer' : l2(kwargs.get(
'weight_decay',
5e-4))}
-
darknet_conv_kwargs[
'padding'] =
'valid'
if kwargs.get(
'strides')==(
2,
2)
else
'same'
-
try:
-
del kwargs[
'weight_decay']
-
except:
-
pass
-
darknet_conv_kwargs.update(kwargs)
-
return Conv2D(*args, **darknet_conv_kwargs)
-
-
#---------------------------------------------------#
-
# 卷积块 -> 卷积 + 标准化 + 激活函数
-
# DarknetConv2D + BatchNormalization + SiLU
-
#---------------------------------------------------#
-
def
DarknetConv2D_BN_SiLU(
*args, **kwargs):
-
no_bias_kwargs = {
'use_bias':
False}
-
no_bias_kwargs.update(kwargs)
-
if
"name"
in kwargs.keys():
-
no_bias_kwargs[
'name'] = kwargs[
'name'] +
'.conv'
-
return compose(
-
DarknetConv2D(*args, **no_bias_kwargs),
-
BatchNormalization(momentum =
0.97, epsilon =
0.001, name = kwargs[
'name'] +
'.bn'),
-
SiLU())
-
-
def
Bottleneck(
x, out_channels, shortcut=True, weight_decay=5e-4, name = ""):
-
y = compose(
-
DarknetConv2D_BN_SiLU(out_channels, (
1,
1), weight_decay=weight_decay, name = name +
'.cv1'),
-
DarknetConv2D_BN_SiLU(out_channels, (
3,
3), weight_decay=weight_decay, name = name +
'.cv2'))(x)
-
if shortcut:
-
y = Add()([x, y])
-
return y
-
-
def
C3(
x, num_filters, num_blocks, shortcut=True, expansion=0.5, weight_decay=5e-4, name=""):
-
hidden_channels =
int(num_filters * expansion)
-
-
x_1 = DarknetConv2D_BN_SiLU(hidden_channels, (
1,
1), weight_decay=weight_decay, name = name +
'.cv1')(x)
-
-
x_2 = DarknetConv2D_BN_SiLU(hidden_channels, (
1,
1), weight_decay=weight_decay, name = name +
'.cv2')(x)
-
for i
in
range(num_blocks):
-
x_1 = Bottleneck(x_1, hidden_channels, shortcut=shortcut, weight_decay=weight_decay, name = name +
'.m.' +
str(i))
-
#----------------------------------------------------------------#
-
route = Concatenate()([x_1, x_2])
-
return DarknetConv2D_BN_SiLU(num_filters, (
1,
1), weight_decay=weight_decay, name = name +
'.cv3')(route)
-
-
def
SPPBottleneck(
x, out_channels, weight_decay=5e-4, name = ""):
-
-
x = DarknetConv2D_BN_SiLU(out_channels //
2, (
1,
1), weight_decay=weight_decay, name = name +
'.cv1')(x)
-
maxpool1 = MaxPooling2D(pool_size=(
5,
5), strides=(
1,
1), padding=
'same')(x)
-
maxpool2 = MaxPooling2D(pool_size=(
9,
9), strides=(
1,
1), padding=
'same')(x)
-
maxpool3 = MaxPooling2D(pool_size=(
13,
13), strides=(
1,
1), padding=
'same')(x)
-
x = Concatenate()([x, maxpool1, maxpool2, maxpool3])
-
x = DarknetConv2D_BN_SiLU(out_channels, (
1,
1), weight_decay=weight_decay, name = name +
'.cv2')(x)
-
return x
-
-
def
resblock_body(
x, num_filters, num_blocks, expansion=0.5, shortcut=True, last=False, weight_decay=5e-4, name = ""):
-
-
-
# 320, 320, 64 => 160, 160, 128
-
x = ZeroPadding2D(((
1,
0),(
1,
0)))(x)
-
x = DarknetConv2D_BN_SiLU(num_filters, (
3,
3), strides = (
2,
2), weight_decay=weight_decay, name = name +
'.0')(x)
-
if last:
-
x = SPPBottleneck(x, num_filters, weight_decay=weight_decay, name = name +
'.1')
-
return C3(x, num_filters, num_blocks, shortcut=shortcut, expansion=expansion, weight_decay=weight_decay, name = name +
'.1'
if
not last
else name +
'.2')
-
-
def
darknet_body(
x, base_channels, base_depth, weight_decay=5e-4):
-
# 640, 640, 3 => 320, 320, 12
-
x = Focus()(x)
-
# 320, 320, 12 => 320, 320, 64
-
x = DarknetConv2D_BN_SiLU(base_channels, (
3,
3), weight_decay=weight_decay, name =
'backbone.stem.conv')(x)
-
# 320, 320, 64 => 160, 160, 128
-
x = resblock_body(x, base_channels *
2, base_depth, weight_decay=weight_decay, name =
'backbone.dark2')
-
# 160, 160, 128 => 80, 80, 256
-
x = resblock_body(x, base_channels *
4, base_depth *
3, weight_decay=weight_decay, name =
'backbone.dark3')
-
feat1 = x
-
# 80, 80, 256 => 40, 40, 512
-
x = resblock_body(x, base_channels *
8, base_depth *
3, weight_decay=weight_decay, name =
'backbone.dark4')
-
feat2 = x
-
# 40, 40, 512 => 20, 20, 1024
-
x = resblock_body(x, base_channels *
16, base_depth, shortcut=
False, last=
True, weight_decay=weight_decay, name =
'backbone.dark5')
-
feat3 = x
-
return feat1,feat2,feat3
yolov5模型：
-
from tensorflow.keras.layers
import (Concatenate, Input, Lambda, UpSampling2D,
-
ZeroPadding2D)
-
from tensorflow.keras.models
import Model
-
-
from nets.CSPdarknet
import (C3, DarknetConv2D, DarknetConv2D_BN_SiLU,
-
darknet_body)
-
from nets.yolo_training
import yolo_loss
-
-
-
#---------------------------------------------------#
-
-
def
yolo_body(
input_shape, anchors_mask, num_classes, phi, weight_decay=5e-4):
-
depth_dict = {
's' :
0.33,
'm' :
0.67,
'l' :
1.00,
'x' :
1.33,}
-
width_dict = {
's' :
0.50,
'm' :
0.75,
'l' :
1.00,
'x' :
1.25,}
-
dep_mul, wid_mul = depth_dict[phi], width_dict[phi]
-
-
base_channels =
int(wid_mul *
64)
# 64
-
base_depth =
max(
round(dep_mul *
3),
1)
# 3
-
-
inputs = Input(input_shape)
-
-
feat1, feat2, feat3 = darknet_body(inputs, base_channels, base_depth, weight_decay)
-
-
P5 = DarknetConv2D_BN_SiLU(
int(base_channels *
8), (
1,
1), weight_decay=weight_decay, name =
'conv_for_feat3')(feat3)
-
P5_upsample = UpSampling2D()(P5)
-
P5_upsample = Concatenate(axis = -
1)([P5_upsample, feat2])
-
P5_upsample = C3(P5_upsample,
int(base_channels *
8), base_depth, shortcut =
False, weight_decay=weight_decay, name =
'conv3_for_upsample1')
-
-
P4 = DarknetConv2D_BN_SiLU(
int(base_channels *
4), (
1,
1), weight_decay=weight_decay, name =
'conv_for_feat2')(P5_upsample)
-
P4_upsample = UpSampling2D()(P4)
-
P4_upsample = Concatenate(axis = -
1)([P4_upsample, feat1])
-
P3_out = C3(P4_upsample,
int(base_channels *
4), base_depth, shortcut =
False, weight_decay=weight_decay, name =
'conv3_for_upsample2')
-
-
P3_downsample = ZeroPadding2D(((
1,
0),(
1,
0)))(P3_out)
-
P3_downsample = DarknetConv2D_BN_SiLU(
int(base_channels *
4), (
3,
3), strides = (
2,
2), weight_decay=weight_decay, name =
'down_sample1')(P3_downsample)
-
P3_downsample = Concatenate(axis = -
1)([P3_downsample, P4])
-
P4_out = C3(P3_downsample,
int(base_channels *
8), base_depth, shortcut =
False, weight_decay=weight_decay, name =
'conv3_for_downsample1')
-
-
P4_downsample = ZeroPadding2D(((
1,
0),(
1,
0)))(P4_out)
-
P4_downsample = DarknetConv2D_BN_SiLU(
int(base_channels *
8), (
3,
3), strides = (
2,
2), weight_decay=weight_decay, name =
'down_sample2')(P4_downsample)
-
P4_downsample = Concatenate(axis = -
1)([P4_downsample, P5])
-
P5_out = C3(P4_downsample,
int(base_channels *
16), base_depth, shortcut =
False, weight_decay=weight_decay, name =
'conv3_for_downsample2')
-
-
out2 = DarknetConv2D(
len(anchors_mask[
2]) * (
5 + num_classes), (
1,
1), strides = (
1,
1), weight_decay=weight_decay, name =
'yolo_head_P3')(P3_out)
-
out1 = DarknetConv2D(
len(anchors_mask[
1]) * (
5 + num_classes), (
1,
1), strides = (
1,
1), weight_decay=weight_decay, name =
'yolo_head_P4')(P4_out)
-
out0 = DarknetConv2D(
len(anchors_mask[
0]) * (
5 + num_classes), (
1,
1), strides = (
1,
1), weight_decay=weight_decay, name =
'yolo_head_P5')(P5_out)
-
return Model(inputs, [out0, out1, out2])
-
-
def
get_train_model(
model_body, input_shape, num_classes, anchors, anchors_mask, label_smoothing):
-
y_true = [Input(shape = (input_shape[
0] // {
0:
32,
1:
16,
2:
8}[l], input_shape[
1] // {
0:
32,
1:
16,
2:
8}[l], \
-
len(anchors_mask[l]), num_classes +
5))
for l
in
range(
len(anchors_mask))]
-
model_loss = Lambda(
-
yolo_loss,
-
output_shape = (
1, ),
-
name =
'yolo_loss',
-
arguments = {
-
'input_shape' : input_shape,
-
'anchors' : anchors,
-
'anchors_mask' : anchors_mask,
-
'num_classes' : num_classes,
-
'label_smoothing' : label_smoothing,
-
'balance' : [
0.4,
1.0,
4],
-
'box_ratio' :
0.05,
-
'obj_ratio' :
1 * (input_shape[
0] * input_shape[
1]) / (
640 **
2),
-
'cls_ratio' :
0.5 * (num_classes /
80)
-
}
-
)([*model_body.output, *y_true])
-
model = Model([model_body.
input, *y_true], model_loss)
-
return model
4. 模型性能测试
模型测试的mAP值：
获取mAP的程序：
-
import os
-
import xml.etree.ElementTree
as ET
-
-
import tensorflow
as tf
-
from PIL
import Image
-
from tqdm
import tqdm
-
-
from utils.utils
import get_classes
-
from utils.utils_map
import get_coco_map, get_map
-
from yolo
import YOLO
-
-
gpus = tf.config.experimental.list_physical_devices(device_type=
'GPU')
-
for gpu
in gpus:
-
tf.config.experimental.set_memory_growth(gpu,
True)
-
-
if __name__ ==
"__main__":
-
-
-
classes_path =
'data/defeat_name.txt'
-
-
MINOVERLAP =
0.5
-
-
VOCdevkit_path =
'VOCdevkit'
-
-
image_ids =
open(os.path.join(VOCdevkit_path,
"VOC2007/ImageSets/Main/test.txt")).read().strip().split()
-
-
if
not os.path.exists(map_out_path):
-
os.makedirs(map_out_path)
-
if
not os.path.exists(os.path.join(map_out_path,
'ground-truth')):
-
os.makedirs(os.path.join(map_out_path,
'ground-truth'))
-
if
not os.path.exists(os.path.join(map_out_path,
'detection-results')):
-
os.makedirs(os.path.join(map_out_path,
'detection-results'))
-
if
not os.path.exists(os.path.join(map_out_path,
'images-optional')):
-
os.makedirs(os.path.join(map_out_path,
'images-optional'))
-
-
class_names, _ = get_classes(classes_path)
-
-
if map_mode ==
0 :
-
print(
"Load model.")
-
yolo = YOLO(confidence =
0.001, nms_iou =
0.5)
-
print(
"Load model done.")
-
-
print(
"Get predict result.")
-
for image_id
in tqdm(image_ids):
-
image_path = os.path.join(VOCdevkit_path,
"VOC2007/JPEGImages/"+image_id+
".jpg")
-
image = Image.
open(image_path)
-
if map_vis:
-
image.save(os.path.join(map_out_path,
"images-optional/" + image_id +
".jpg"))
-
yolo.get_map_txt(image_id, image, class_names, map_out_path)
-
print(
"Get predict result done.")
-
-
-
实例预测：
转载：https://blog.csdn.net/ncusz/article/details/128472557
查看评论