.png){kind=icon}
.png){kind=icon}
Return
Sweet potato grasping point recognition in complex backgrounds: a method based on salient object detection
.png){kind=icon}
PRE
AI
DOI：10.15302/J-FASE-2025653.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address challenges in crop grasping tasks for agricultural robots, specifically, poor crop background segmentation and limited adaptability in grasp point localization, this paper proposes a saliency guided segmentation approach. This method improves both object recognition and grasp point detection, thereby optimizing robot grasping performance and increasing success rates, even under complex environmental conditions. The proposed network uses a boundary aware detection strategy built on an encoder decoder architecture with an improvement module. First, standard convolutions are replaced by dynamic convolution to improve feature representation. Second, a Haar wavelet downsampling module is introduced to improve multi scale feature extraction. Finally, the standard squeeze and excitation attention block is improved with edge enhancement, which is embedded at each decoding stage to emphasize boundary information. In benchmark tests, the proposed model achieved a mean absolute error of 10.9%, with F-, E- and S-measures of 97.0%, 98.4%, and 96.8%, respectively. When deployed on an agricultural robot platform, it achieved a 78.0% grasping success rate, processing images at 35 frames per second. These results demonstrate that the proposed network reliably identifies and localizes optimal grasp points under real world conditions.
Keywords:
Agricultural robots
edge detection
salient object detection
sweet potato
Journal
IF:
2.8
Papers:
339
Citations:
1.1K