Research on Soybean Seedling Stage Recognition Based on Swin Transformer DOI Creative Commons

Kai Ma,

Jinkai Qiu,

Kang Ye

et al.

Agronomy, Journal Year: 2024, Volume and Issue: 14(11), P. 2614 - 2614

Published: Nov. 6, 2024

Accurate identification of the second and third compound leaf periods soybean seedlings is a prerequisite to ensure that soybeans are chemically weeded after seedling at optimal application period. period susceptible natural light complex field background factors. A transfer learning-based Swin-T (Swin Transformer) network proposed recognize different stages stage. drone was used collect images true stage, first data enhancement methods such as image rotation brightness were expand dataset, simulate drone’s collection shooting angles weather conditions, enhance adaptability model. The environment equipment directly affect quality captured images, in order test anti-interference ability models, Gaussian blur method set degrees. model optimized by introducing learning combining hyperparameter combination experiments optimizer selection experiments. performance compared with MobileNetV2, ResNet50, AlexNet, GoogleNet, VGG16Net models. results show has an average accuracy 98.38% set, which improvement 11.25%, 12.62%, 10.75%, 1.00%, 0.63% respectively. best terms recall F1 score. In degradation motion level model, maximum accuracy, overall index, index 87.77%, 6.54%, 2.18%, 7.02%, 7.48%, 10.15%, 3.56%, 2.5% higher than fuzzy 94.3%, 3.85%, 1.285%, Compared 12.13%, 15.98%, 16.7%, 2.2%, 1.5% higher, Taking into account various indicators, can still maintain high recognition demonstrate good even when inputting blurry caused interference shooting. It meet growth environments, providing basis for post-seedling chemical weed control during soybeans.

Language: Английский

Aerial Systems for Releasing Natural Enemy Insects of Purple Loosestrife Using Drones DOI Creative Commons
Kushal Naharki, Christopher J. Hayes, Yong‐Lak Park

et al.

Drones, Journal Year: 2024, Volume and Issue: 8(11), P. 635 - 635

Published: Nov. 1, 2024

Lythrum salicaria (purple loosestrife) is an invasive species that displaces native wetland flora in the USA. The detection and manual release of biological control agents for L. challenging because inhabits many inaccessible areas. This study was conducted to develop aerial systems its natural enemy, Galerucella calmariensis (Coleoptera: Chrysomelidae). We determined optimal sensors flight height designed deployment method G. calmariensis. Drone-based surveys were at various heights utilizing RGB, multispectral, thermal sensors. also developed insect container (i.e., bug ball) Our findings indicated flowers detectable with RGB sensor ≤ 15 m above canopy. post-release mortality feeding efficiency did not significantly differ from group (non-aerial release), indicating feasibility targeted innovative establishes a critical foundation future development sophisticated automated plants precise agents, advancing ecological management conservation efforts.

Language: Английский

Citations

0
Research on Soybean Seedling Stage Recognition Based on Swin Transformer DOI Creative Commons

Kai Ma,

Jinkai Qiu,

Kang Ye

et al.

Agronomy, Journal Year: 2024, Volume and Issue: 14(11), P. 2614 - 2614

Published: Nov. 6, 2024

Accurate identification of the second and third compound leaf periods soybean seedlings is a prerequisite to ensure that soybeans are chemically weeded after seedling at optimal application period. period susceptible natural light complex field background factors. A transfer learning-based Swin-T (Swin Transformer) network proposed recognize different stages stage. drone was used collect images true stage, first data enhancement methods such as image rotation brightness were expand dataset, simulate drone’s collection shooting angles weather conditions, enhance adaptability model. The environment equipment directly affect quality captured images, in order test anti-interference ability models, Gaussian blur method set degrees. model optimized by introducing learning combining hyperparameter combination experiments optimizer selection experiments. performance compared with MobileNetV2, ResNet50, AlexNet, GoogleNet, VGG16Net models. results show has an average accuracy 98.38% set, which improvement 11.25%, 12.62%, 10.75%, 1.00%, 0.63% respectively. best terms recall F1 score. In degradation motion level model, maximum accuracy, overall index, index 87.77%, 6.54%, 2.18%, 7.02%, 7.48%, 10.15%, 3.56%, 2.5% higher than fuzzy 94.3%, 3.85%, 1.285%, Compared 12.13%, 15.98%, 16.7%, 2.2%, 1.5% higher, Taking into account various indicators, can still maintain high recognition demonstrate good even when inputting blurry caused interference shooting. It meet growth environments, providing basis for post-seedling chemical weed control during soybeans.

Language: Английский

Citations

0