arrow
Volume 14, Issue 4
Insect real-time detection in complex environments based on improved YOLOV3

Juan Wang

J. Info. Comput. Sci. , 14 (2019), pp. 291-299.

Export citation
  • Abstract
1 College of Information and Science Technology, Jinan University, Guangzhou, 51000, China (Received November 02 2019, accepted December 28 2019) The combination of advanced computer vision technology and insect image recognition technology can be effectively applied to environmental monitoring, pest diagnosis, epidemiology and other fields. However, accurate location and classification of relatively small insects in complex scenes has always been a difficult problem for this technology. Although YOLOv3 combines deep features with shallow features to facilitate the detection of small objects, experiments have found that YOLOv3 has more undetected cases for small target insects in complex backgrounds. In order to solve this problem, this paper optimizes YOLOv3 algorithm. Firstly, the SE blocks are embedded into YOLOv3 network to learn global features and enhance the expression ability of feature maps, so that the network can detect more objects. Because YOLOv3 itself has a complex network structure and the amount of parameters and calculations are increased after embedding the SE block, so this paper also uses depthwise separable convolutions, which greatly reduces the amount of parameters and computation under the condition of little loss of accuracy, thus improving the detection speed. Training and testing on the insect dataset made in this paper, the original YOLOv3 runs at 33 f / s, and the mean Average Precision (mAP) is only 86.8%, While the improved YOLOv3 runs at 38 f / s, the mAP reaches 90.6%. The improved algorithm can detect more targets, reduce the omission factor and improve the detection speed.
  • AMS Subject Headings

  • Copyright

COPYRIGHT: © Global Science Press

  • Email address
  • BibTex
  • RIS
  • TXT
@Article{JICS-14-291, author = {Juan Wang}, title = {Insect real-time detection in complex environments based on improved YOLOV3}, journal = {Journal of Information and Computing Science}, year = {2024}, volume = {14}, number = {4}, pages = {291--299}, abstract = {1 College of Information and Science Technology, Jinan University, Guangzhou, 51000, China (Received November 02 2019, accepted December 28 2019) The combination of advanced computer vision technology and insect image recognition technology can be effectively applied to environmental monitoring, pest diagnosis, epidemiology and other fields. However, accurate location and classification of relatively small insects in complex scenes has always been a difficult problem for this technology. Although YOLOv3 combines deep features with shallow features to facilitate the detection of small objects, experiments have found that YOLOv3 has more undetected cases for small target insects in complex backgrounds. In order to solve this problem, this paper optimizes YOLOv3 algorithm. Firstly, the SE blocks are embedded into YOLOv3 network to learn global features and enhance the expression ability of feature maps, so that the network can detect more objects. Because YOLOv3 itself has a complex network structure and the amount of parameters and calculations are increased after embedding the SE block, so this paper also uses depthwise separable convolutions, which greatly reduces the amount of parameters and computation under the condition of little loss of accuracy, thus improving the detection speed. Training and testing on the insect dataset made in this paper, the original YOLOv3 runs at 33 f / s, and the mean Average Precision (mAP) is only 86.8%, While the improved YOLOv3 runs at 38 f / s, the mAP reaches 90.6%. The improved algorithm can detect more targets, reduce the omission factor and improve the detection speed. }, issn = {1746-7659}, doi = {https://doi.org/}, url = {http://global-sci.org/intro/article_detail/jics/22405.html} }
TY - JOUR T1 - Insect real-time detection in complex environments based on improved YOLOV3 AU - Juan Wang JO - Journal of Information and Computing Science VL - 4 SP - 291 EP - 299 PY - 2024 DA - 2024/01 SN - 14 DO - http://doi.org/ UR - https://global-sci.org/intro/article_detail/jics/22405.html KW - target detection, real-time detection, convolution neural network, YOLOv3. AB - 1 College of Information and Science Technology, Jinan University, Guangzhou, 51000, China (Received November 02 2019, accepted December 28 2019) The combination of advanced computer vision technology and insect image recognition technology can be effectively applied to environmental monitoring, pest diagnosis, epidemiology and other fields. However, accurate location and classification of relatively small insects in complex scenes has always been a difficult problem for this technology. Although YOLOv3 combines deep features with shallow features to facilitate the detection of small objects, experiments have found that YOLOv3 has more undetected cases for small target insects in complex backgrounds. In order to solve this problem, this paper optimizes YOLOv3 algorithm. Firstly, the SE blocks are embedded into YOLOv3 network to learn global features and enhance the expression ability of feature maps, so that the network can detect more objects. Because YOLOv3 itself has a complex network structure and the amount of parameters and calculations are increased after embedding the SE block, so this paper also uses depthwise separable convolutions, which greatly reduces the amount of parameters and computation under the condition of little loss of accuracy, thus improving the detection speed. Training and testing on the insect dataset made in this paper, the original YOLOv3 runs at 33 f / s, and the mean Average Precision (mAP) is only 86.8%, While the improved YOLOv3 runs at 38 f / s, the mAP reaches 90.6%. The improved algorithm can detect more targets, reduce the omission factor and improve the detection speed.
Juan Wang. (2024). Insect real-time detection in complex environments based on improved YOLOV3. Journal of Information and Computing Science. 14 (4). 291-299. doi:
Copy to clipboard
The citation has been copied to your clipboard