Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 9, 1216(2022)
Forestry pest detection optimization based on deep learning
Figures & Tables(17)
Fig. 11. Comparison of Pest-YOLOv4 and YOLOv4 on some test pictures
Table 1. Statistics of insects in the training set
View tableView in ArticleTable 1. Statistics of insects in the training set
昆虫名 样本数量 红脂大小蠹（Leconte） 2 216 松十二齿小蠹（Boerner） 1 595 华山松大小蠹（Armandi） 1 765 四眼小蠹（Coleoptera） 2 091 六齿小蠹（Acuminatus） 953 云杉八齿小蠹（Linnaeus） 1 727
Table 2. Experimental parameters setting
View tableView in ArticleTable 2. Experimental parameters setting
参数 值 批大小 2 训练次数 250 动量 0.9 前100个Epoch学习率 0.01 100 Epoch后学习率 0.001
Table 3. Comparison of anchor box optimization
View tableView in ArticleTable 3. Comparison of anchor box optimization
网络模型 先验框聚类策略 Precision/% Recall/% mAP/% YOLOv4 K-means 60.3 88.4 86.2 YOLOv4 K-means++ 60.6 88.9 86.7
Table 4. Comparison of five models for forestry pest detection
View tableView in ArticleTable 4. Comparison of five models for forestry pest detection
网络模型 Precision/
%
Recall/% FPS Faster R-CNN（ResNet101） 63.0 92.4 4.6 SSD 57.2 83.9 29.3 YOLOv3 58.4 85.7 34.5 YOLOv4 60.3 88.4 39.2 Pest-YOLOv4 67.5 92.8 33.4
Table 5. Detection results of Pest-YOLOv4 and YOLOv4
View tableView in ArticleTable 5. Detection results of Pest-YOLOv4 and YOLOv4
昆虫名称 YOLOv4 Pest-YOLOv4 Precision/% Recall/% AP/% Precision/% Recall/% AP/% Leconte 71.5 94.0 95.4 81.9 95.3 96.2 Boerner 57.5 89.1 88.9 81.3 96.1 94.5 Armandi 65.9 86.2 84.8 68.6 93.4 87.6 Coleoptera 39.2 87.2 80.5 36.5 95.4 86.2 Acuminatus 52.5 82.9 79.8 51.2 94.9 87.8 Linnaeus 75.4 91.1 88.0 85.3 81.9 90.3 均值 60.3 88.4 86.2 67.5 92.8 90.4
Table 6. Detection effects of various improvement strategies
View tableView in ArticleTable 6. Detection effects of various improvement strategies
网络模型 改进策略 Precision/% Recall/% mAP/% K-means++ ECA-CBAM SPP-PANet Focal Loss YOLOv4 60.3 88.4 86.2 YOLOv4-A √ √ 58.8 92.6 88.5 YOLOv4-B √ √ 60.9 90.9 87.1 YOLOv4-C √ √ 61.6 91.7 88.3 YOLOv4-D √ √ √ 63.5 91.9 89.0 Pest-YOLOv4 √ √ √ √ 67.5 92.8 90.4
Tools
Get Citation
Copy Citation Text
Yan ZHAO, Ying-an LIU, Qiao-lin YE, Xiao-liang ZHOU. Forestry pest detection optimization based on deep learning[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(9): 1216
Paper Information
Category: Research Articles
Received: Mar. 8, 2022
Accepted: --
Published Online: Sep. 14, 2022
The Author Email: Ying-an LIU (lyastat@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20