Fruit Detection in the Wild: The Impact of Varying Conditions and Cultivar

Halstead, Michael; Denman, Simon; Fookes, Clinton; McCool, Chris

doi:10.1109/dicta51227.2020.9363407

Cited by 25 publications

(33 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bargoti and Underwood (2017) outlined the benefits of convolutional neural networks (CNN) for fruit segmentation. More recently, the potential to perform fruit detection in the wild was explored by Halstead et al (2020). It was shown that impressive performance for fruit detection could be achieved in vastly different fields by leveraging multi-task learning.…”

Section: Vision In Agriculturementioning

confidence: 99%

“…The sweet pepper dataset was captured at the University of Bonn's campus Klein-Altendorf (CKA) in a commercial glasshouse. The dataset was captured by Smitt et al (2021) on PATHoBot under similar conditions to that captured by Halstead et al (2020). Figure 2 (top row) outlines two examples of the RGB and masks from the BUP20 dataset.…”

Section: Sweet Pepper Dataset (Bup20)mentioning

confidence: 99%

“…We enhance our previous work in Halstead et al (2018Halstead et al ( , 2020, where the quality (ripeness) was introduced as a parallel For SB20, N = 8 species of plants and for BUP20, N = 5 ripeness estimates (red, yellow, green, mixed-red, and mixedyellow). We use this approach as our previous work (Halstead et al, 2018(Halstead et al, , 2020 demonstrated that introducing a single superclass provided superior performance for the detection of sweet pepper (compared to N super-classes). This was attributed to the fact that all samples could be used in a single network structure to derive a strong generalized super-class classifier to detect the presence of sweet pepper (BUP20).…”

Section: Instance-based Semantic Segmentationmentioning

confidence: 99%

“…The parallel structure ensures both general (super) and specific (sub) classification information is learned. Halstead et al (2018Halstead et al ( , 2020 showed that using this parallel layer had considerable advantages when applied to sweet pepper detection (super-class) and ripeness estimation (subclass). We demonstrate, for the first time, that this parallel layer can also be applied to plants in the field to perform plant detection (super-class) and species (sub-class) classification.…”

Section: Instance-based Semantic Segmentationmentioning

confidence: 99%

“…In this article, we present a platform agnostic algorithm for monitoring the state of arable farmland (crop/weed management) and glasshouses (crop management). We propose to improve the AI and tracking components outlined in Halstead et al (2018Halstead et al ( , 2020 and evaluate it on both PATHoBot (Smitt et al, 2021) (glasshouse) and BonnBot-I (Ahmadi et al, 2021a) (arable farmland). This approach monitors the crops in the various environments and assists the farmer in making informed management decisions.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Crop Agnostic Monitoring Driven by Deep Learning

et al. 2021

Self Cite

View full text Add to dashboard Cite

Farmers require diverse and complex information to make agronomical decisions about crop management including intervention tasks. Generally, this information is gathered by farmers traversing their fields or glasshouses which is often a time consuming and potentially expensive process. In recent years, robotic platforms have gained significant traction due to advances in artificial intelligence. However, these platforms are usually tied to one setting (such as arable farmland), or algorithms are designed for a single platform. This creates a significant gap between available technology and farmer requirements. We propose a novel field agnostic monitoring technique that is able to operate on two different robots, in arable farmland or a glasshouse (horticultural setting). Instance segmentation forms the backbone of this approach from which object location and class, object area, and yield information can be obtained. In arable farmland, our segmentation network is able to estimate crop and weed at a species level and in a glasshouse we are able to estimate the sweet pepper and their ripeness. For yield information, we introduce a novel matching criterion that removes the pixel-wise constraints of previous versions. This approach is able to accurately estimate the number of fruit (sweet pepper) in a glasshouse with a normalized absolute error of 4.7% and an R2 of 0.901 with the visual ground truth. When applied to cluttered arable farmland scenes it improves on the prior approach by 50%. Finally, a qualitative analysis shows the validity of this agnostic monitoring algorithm by supplying decision enabling information to the farmer such as the impact of a low level weeding intervention scheme.

show abstract

Section: Vision In Agriculturementioning

confidence: 99%

Section: Sweet Pepper Dataset (Bup20)mentioning

confidence: 99%

Section: Instance-based Semantic Segmentationmentioning

confidence: 99%

Section: Instance-based Semantic Segmentationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Crop Agnostic Monitoring Driven by Deep Learning

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

GrowliFlower: An image time‐series dataset for GROWth analysis of cauLIFLOWER

Kierdorf¹,

Junker‐Frohn

Delaney³

et al. 2022

Journal of Field Robotics

View full text Add to dashboard Cite

In this paper, we present GrowliFlower, a georeferenced, image-based unmanned aerial vehicle time-series dataset of two monitored cauliflower fields (0.39 and 0.60 ha) acquired in 2 years, 2020 and 2021. The proposed dataset contains RGB and multispectral orthophotos with coordinates of approximately 14,000 individual cauliflower plants. The coordinates enable the extraction of complete and incomplete time-series of image patches showing individual plants. The dataset contains the collected phenotypic traits of 740 plants, including the developmental stage and plant and cauliflower size. The harvestable product is completely covered by leaves, thus, plant IDs and coordinates are provided to extract image pairs of plants pre-and postdefoliation. In addition, to facilitate classification, detection, segmentation, instance segmentation, and other similar computer vision tasks, the proposed dataset contains pixel-accurate leaf and plant instance segmentations, as well as stem annotations. The proposed dataset was created to facilitate the development and evaluation of various machine-learning approaches. It focuses on the analysis of growth and development of cauliflower and the derivation of phenotypic traits to advance automation in agriculture.Two baseline results of instance segmentation tasks at the plant and leaf level based on labeled instance segmentation data are presented. The complete GrowliFlower dataset is publicly available (http://rs.ipb.uni-bonn.de/data/growliflower/).

show abstract