Abstract
In the pollution remediation operation, it is of great significance for the work of
pollution remediation to effectively guarantee the personal safety of workers in toxic
environment. Based on the deep learning technology, this paper explores the
personnel dress code and behavior detection method, and detects whether the
personnel dress code and behavior are abnormal in real time, so as to improve the
safety protection level of the staff in the pollution remediation environment. The main
work content is as follows:
(1) This subject analysis and comparison of traditional target detection algorithm
and target detection algorithm based on depth study, the experimental results show
that the traditional types of target detection algorithm is relatively single and greatly
influenced by environmental factors, so the target detection algorithm based on deep
learning is more suitable for complex pollution in this paper, we study the repair work
environment. In this paper, a deep learning method is adopted to study the personnel
dress code and behavior detection in pollution remediation environment.
(2) Based on improved Faster R-CNN algorithm, the method of detecting
personnel dress code was studied. Made the personnel environment pollution repair
operations Dress Code dataset, contrast research Faster R-CNN algorithm commonly
used backbone network and the division of the data sets in different ratio, optimal
model, combining with the super parameter adjustment on the basis of regression
tasks by optimizing algorithm of the objective function, experiments show that the
final mAP value 1.06% increased and single image detection speed of 0.02 s.
(3) Based on YOLOv3 algorithm, the method of personnel behavior detection is
studied. Bahavior Dataset was made based on the status level of human behavior. The
commonly used backbone network in YOLOv3 algorithm is used to divide different
proportions of data sets, and the optimal model is obtained by combining with the
adjustment of super parameters. The experiment shows that the mAP value can reach
95.40%.
(4) A real-time monitoring device for the toxic gas content in the pollution
remediation environment is designed. Aiming at the problem that the cause of
behavior cannot be determined when the behavior detection may be missed or wrong
and the severity is "medium", the results of the toxic gas monitoring module are
万方数据
