Table 1.

Public weed image datasets and their features.

Reference	Datasets	Purpose	Plant	Image Size	Features
[18]	Perennial ryegrass and weed	Weed detection and control	Dandelion, ground ivy, spotted spurge, and ryegrass	1920 × 1080 33,086	It includes 17,600 positive images (contain target weeds) and 15,486 negative images (contain perennial ryegrass with no target weeds).
[19]	Grass-Broadleaf	Weed detection by using ConvNets	Soil, soybean, broadleaf, and grass weeds	4000 × 3000 15,336	Data are from a set of images captured using a UAV and the SLIC algorithm. These images are segmented, and the segments are annotated manually. The ratio of soil: soybeans: grass: broadleaf weeds is roughly 3:7:3:1 (Figure 1a).
[20]	Plant seedlings dataset	Identifying plant species and weeding in the early growth stage	12 weed and crop species of Danish arable land	5184 × 3456 407	Each image is provided with an ID and associated with a single species. The dataset contains a full image, automatically segmented plants, and single plants that are not segmented.
[21]	DeepWeeds	Classification of multiple weed species based on deep learning	8 nationally significant weed species native to 8 locations across northern Australia	256 × 256 17,509	Each class contains between 1009 and 1125 images of the corresponding species, with a total of over 8000 images of positive species classes.
[22]	Open Plant Phenotype Database	Plant detection and classification algorithms	47 species of common weeds in Denmark	1000 × 1000 7590	It includes 47 different species of monocotyledonous and dicotyledonous weeds in arable crops in Denmark. Several plant species were cultivated in a semifield setting to mimic natural growth conditions.
[23]	WeedNet	Dense semantic classification, vegetation detection	Crops and weeds	/ 465	Three kinds of multispectral image datasets are included: one contains only 132 images of crops, the other has 243 images of weeds, and the third one contains 90 images of crop–weed.
[24]	Sugar beet	Plant classification, localization, and mapping	Sugar beets and 9 different types of weed	1296 × 966 >10,000	Data were recorded 3 times per week until the field was no longer accessible to the machinery without damaging the crops. The robot carried a four-channel multispectral camera and an RGB-D sensor.
[25]	Rice seedlings and weeds	Image segmentation of rice seedling and weeds	Rice seedlings and weed background	912 × 1024 224	The images were selected in the paddy fields, and all weeds were in early growth stages. The data sample included GT and RGB images (Figure 1c).
[26]	Food crops and weed	Crop and weed identification	6 food crops and 8 weed species	720 × 1280 1118	Datasets of 14 basic food crops and weeds in controlled environment and field conditions at different growth stages and manually annotated images are included (Figure 1d).
[27]	Crop and weed	Instance segmentation for fine detection	Maize, the common bean, and a variety of weeds	1200 × 2048 2489	The crops include maize and the common bean. Weeds include cultivated and natural weeds. Each mask is annotated with the species name of the plant.
[28]	Flavia	Plant leaf classification	Leaves of 32 plants	1600 × 1200 1907	Each plant has a minimum of 50 leaves and a maximum of 77. The background of the leaf image is white (Figure 1b).
[29]	CropDeep	Crop classification and testing	30 common vegetables and fruits	1000 × 1000 31,147	At least 1100 annotated samples per category and vegetables or fruits with different parts and periods of growth are included. A high degree of similarity exists among certain categories in the dataset.