Research on Automatic Monitoring Technology of Combination of Image Processing and Deep Learning

Main Article Content

Zheng Li

Abstract

Recently, the equipment automatically monitors whether the electricity consumption is normal or not is an
extremely important link in the system. The safe operation of the system is related to the safety of life and property.
This has caused the number of equipment to monitor the status to continue to grow, resulting in shortages of
inspection workers and personnel on duty, increased operating costs, and low labor efficiency. The equipment is
generally built in areas with more complex environments, and it is not safe for staff to perform inspections.
Moreover, with the development of computer software and hardware, deep learning technology has gradually
matured. Researchers have used deep learning technology in various fields, including the field of automatic
equipment monitoring, so that workers do not need to manually copy the status of equipment on site, and can
directly use the front end The device automatically monitors the status of the device. However, the use of deep
learning methods cannot simultaneously identify various types of objects on the equipment and determine their
status, and when the shooting position of the inspection device is deviated or the environment is bad, the
recognition of the object category will be affected, thereby the state of the object The discrimination produces a
large error. This paper analyzes the problems encountered by the equipment automatic monitoring technology in
view of the above problems, and proposes a method combining image processing and deep learning according to
the characteristics of the equipment scene, and uses this method to analyze and process the inspection logo. First,
use the deep learning label detection method to identify the category of the object. For this part, this paper
proposes a multi-label detection model, which is based on the original YOLOv3 model and optimizes the FPN
feature fusion layer and loss function part , And use the K-means clustering method to modify the size of the a
priori box of the convolutional neural network to fit the size of the object in the power equipment scene; secondly,
according to different categories, different image processing methods are used to determine the state. For this part,
this article uses The multi-category comprehensive discrimination method judges the state of each object. This
method integrates various image processing methods according to different categories after the oral logo is
detected. The problem of judging the state of change by fans. Finally, the state discrimination algorithm is
integrated with the deep learning logo detection network to form an automatic equipment monitoring technology
that combines image processing and deep learning.

Article Details

How to Cite
Zheng Li. (2021). Research on Automatic Monitoring Technology of Combination of Image Processing and Deep Learning. CONVERTER, 2021(7), 1007-1017. Retrieved from http://converter-magazine.info/index.php/converter/article/view/589
Section
Articles