Search

Abstract In agricultural production activities, the growth of crops always accompanies the competition of weeds for nutrients and sunlight. In order to mitigate the adverse effects of weeds on yield, we apply semantic segmentation techniques to differentiate between seedlings and weeds, leading to precision weeding. The proposed EPAnet employs a loss function coupled with Cross-entropy loss and Dice loss to enhance attention to feature information. A multi-Decoder cooperative module based on ERFnet is designed to enhance information transfer during feature mapping. The SimAM is introduced to enhance position recognition. DO-CONV is used to replace the traditional convolution Feature Pyramid Networks (FPN) connection layer to integrate feature information, improving the model’s performance on leaf edge processing, and is named FDPN. Moreover, the Overall Accuracy has been improved by 0.65%, the mean Intersection over Union (mIoU) by 1.91%, and the Frequency-Weighted Intersection over Union (FWIoU) by 1.19%. Compared to other advanced methods, EPAnet demonstrates superior image segmentation results in complex natural environments with uneven lighting, leaf interference, and shadows.

Bean crop and weed detection is a key component of precision agriculture, which can distinguish between bean crop and weeds. Accurate identification of weeds is essential for precision weed management. This study introduces PF-UperNet, a semantic segmentation approach rooted in an encoder-decoder architecture, designed to autonomously distinguish between bean crop and weeds using advanced computer vision techniques. Our methodology refines the foundational UperNet in several significant ways: Firstly, we adopt the PoolFormer-S12 as a substitute for UperNet’s backbone structure, aiming to reduce the model parameters and boost its performance metrics. Secondly, the Efficient Channel Attention (ECA) mechanism is integrated into both the PoolFormer-S12 and the Decoder, sharpening the network’s focus on extracting salient channel features. Then, within the Decoder, the Feature Alignment Pyramid Network (FaPN) supplants the conventional Feature Pyramid Network (FPN) module, remedying the misalignment issues observed in UperNet’s FPN feature maps. Lastly, we replace the Cross-Entropy loss with a combination of Cross-Entropy loss and Dice coefficient loss to increase the model’s attention on regions to be detected. Empirical evidence underlines the efficacy of our technique, with a Mean Intersection over Union (MIoU) of 87.45%, a Mean Pixel Accuracy (MPA) of 96.82%, and a total of 46.16M parameters encapsulated. Relative to the benchmark UperNet, our model demonstrates enhancements of 1.08% and 0.25% in MIoU and MPA, respectively, and accomplishes a parameter reduction of 27.92%. Experimental results demonstrate that the proposed model achieves remarkable detection performance in terms of MIoU, MPA, and model parameters. It can provide an effective detection method for weed management.