Object pest detection method based on lightweight SSD_RA algorithm DOI
Shixuan Li, Hongxing Peng, Jingqi Yuan

и другие.

Опубликована: Дек. 9, 2022

In order to detect the types and quantities of pests in rice fields quickly accurately, a lightweight target pest detection method SSD_RA based on SSD algorithm is proposed. deal with problems high missed rate, inaccurate positioning, slow speed, large number model parameters low accuracy model, ResNet feature extraction network was introduced optimized. The first prediction layer connected Conv3_x module network, all layers after were dropped. reduced, so that more lightweight, speed improved, redundant features are reduced ensure model. addition, aiming at characteristics small target, structure adjusted, output underlying Conv2_x layer. candidate box each cell new 6, which accurately divides boundaries large, medium boxes. experimental results show mAP improved this paper 84.1%, 23.4 percentage points higher than original reasoning time CPU GPU environment 0.056s 0.009s, 0.101s 0.005s faster size 51.9MB. It about 7/100 Compared other models, 7.8 4.2 EfficientDet RFCN, respectively. effective detecting insect reduces rate.

Язык: Английский

Object pest detection method based on lightweight SSD_RA algorithm DOI
Shixuan Li, Hongxing Peng, Jingqi Yuan

и другие.

Опубликована: Дек. 9, 2022

In order to detect the types and quantities of pests in rice fields quickly accurately, a lightweight target pest detection method SSD_RA based on SSD algorithm is proposed. deal with problems high missed rate, inaccurate positioning, slow speed, large number model parameters low accuracy model, ResNet feature extraction network was introduced optimized. The first prediction layer connected Conv3_x module network, all layers after were dropped. reduced, so that more lightweight, speed improved, redundant features are reduced ensure model. addition, aiming at characteristics small target, structure adjusted, output underlying Conv2_x layer. candidate box each cell new 6, which accurately divides boundaries large, medium boxes. experimental results show mAP improved this paper 84.1%, 23.4 percentage points higher than original reasoning time CPU GPU environment 0.056s 0.009s, 0.101s 0.005s faster size 51.9MB. It about 7/100 Compared other models, 7.8 4.2 EfficientDet RFCN, respectively. effective detecting insect reduces rate.

Язык: Английский

Процитировано

1