暂无摘要。

Weed suppression is an important factor affecting crop yields. Precise identification of weed species will contribute to automatic weeding by applying proper herbicides, hoeing position determination, and hoeing depth to specific plants as well as reducing crop injury. However, the lack of datasets of weeds in the field has limited the application of deep learning techniques in weed management. In this paper, it presented a dataset of weeds in fields, Weed25, which contained 14,035 images of 25 different weed species. Both monocot and dicot weed image resources were included in this dataset. Meanwhile, weed images at different growth stages were also recorded. Several common deep learning detection models-YOLOv3, YOLOv5, and Faster R-CNN-were applied for weed identification model training using this dataset. The results showed that the average accuracy of detection under the same training parameters were 91.8%, 92.4%, and 92.15% respectively. It presented that Weed25 could be a potential effective training resource for further development of in-field real-time weed identification models. The dataset is available at https://pan.baidu.com/s/1rnUoDm7IxxmX1n1LmtXNXw; the password is rn5h.

BACKGROUND: Amaranthus palmeri is an aggressive and prolific weed species with major impact on agricultural yield and is a prohibited noxious weed across the Midwest. Morphological identification of A. palmeri from other Amaranthus species is extremely difficult in seeds, which has led to genetic testing for seed identification in commercial seed lots.
RESULTS: We created an inexpensive and reliable genetic test based on novel, species-specific, single nucleotide polymorphisms (SNPs) from GBS (Genotyping by Sequencing) data. We report three SNP-based genetic tests for identifying A. palmeri alone or in a mixed pool of Amaranthus spp. Sensitivity ranged from 99.8 to 100%, specificity from 99.59 to 100%. Accuracy for all three tests is > 99.7%. All three are capable of reliably detecting one A. palmeri seed in a pool of 200 Amaranthus spp. seeds. The test was validated across 20 populations of A. palmeri, along with eight other Amaranthus species, the largest and most genetically diverse panel of Amaranthus samples to date.
CONCLUSIONS: Our work represents a marked improvement over existing commercial assays resulting in an identification assay that is (i) accurate, (ii) robust, (iii) easy to interpret and (iv) applicable to both leaf tissue and pools of up to 200 seeds. Included is a data transformation method for calling of closely grouped competitive fluorescence assays. We also present a comprehensive GBS dataset from the largest geographic panel of Amaranthus populations sequenced. Our approach serves as a model for developing markers for other difficult to identify species. © 2021 Society of Chemical Industry.

An image processing algorithm for detecting individual weeds was developed and evaluated. Weed detection processes included were normalized excessive green conversion, statistical threshold value estimation, adaptive image segmentation, median filter, morphological feature calculation and Artificial Neural Network (ANN). The developed algorithm was validated for its ability to identify and detect weeds and crop plants under uncontrolled outdoor illuminations. A machine vision implementing field robot captured field images under outdoor illuminations and the image processing algorithm automatically processed them without manual adjustment. The errors of the algorithm, when processing 666 field images, ranged from 2.1 to 2.9%. The ANN correctly detected 72.6% of crop plants from the identified plants, and considered the rest as weeds. However, the ANN identification rates for crop plants were improved up to 95.1% by addressing the error sources in the algorithm. The developed weed detection and image processing algorithm provides a novel method to identify plants against soil background under the uncontrolled outdoor illuminations, and to differentiate weeds from crop plants. Thus, the proposed new machine vision and processing algorithm may be useful for outdoor applications including plant specific direct applications (PSDA).