
PLoS ONE, Journal Year: 2024, Volume and Issue: 19(1), P. e0296314 - e0296314
Published: Jan. 5, 2024
The development of automated grading equipment requires achieving high throughput and precise detection disease spots on jujubes. However, the current algorithms are inadequate in accomplishing these objectives due to their density, varying sizes shapes, limited location information regarding This paper proposes a method called JujubeSSD, boost precision identifying jujubes based single shot multi-box detector (SSD) network. In this study, diverse dataset comprising varied densities, multiple details was created through artificial collection data augmentation. parameter obtained from transfer learning into backbone feature extraction network SSD model, which reduced time spot 0.14 s. To enhance target detail features improve recognition weak information, traditional convolution layer replaced with deformable convolutional networks (DCNs). Furthermore, address challenge shapes regions jujubes, path aggregation pyramid (PAFPN) balanced (BFP) were integrated Experimental results demonstrate that mean average at IoU (intersection over union) threshold 0.5 ( [email protected] ) JujubeSSD reached 97.1%, representing an improvement approximately 6.35% compared original algorithm. When existing algorithms, such as YOLOv5 Faster R-CNN, improvements 16.84% 8.61%, respectively. Therefore, proposed for detecting jujube achieves superior performance surface meets requirements practical application agricultural production.
Language: Английский