Development of maize plant dataset for intelligent recognition and weed control

This paper focuses on the development of maize plant datasets for the purposes of recognizing maize plants and weed species, as well as the precise automated application of herbicides to the weeds. The dataset includes 36,374 images captured with a high-resolution digital camera during the weed survey and 500 images annotated with the Labelmg suite. Images of the eighteen farmland locations in North Central Nigeria, containing the maize plants and their associated weeds were captured using a high-resolution camera in each location. This dataset will serve as a benchmark for computer vision and machine learning tasks in the intelligent maize and weed recognition research.


Specifications
Engineering Specific subject area Computing Vision applied to annotated maize and weed images critical for development of computer-based autonomous systems that reduce human involvement in crop production activities, especially maize recognition, weed identification and control in a maize farmland. Type of data Image How the data were acquired The raw wet and dry seasons images have been captured using a high-resolution digital camera during the weed survey. Specifically, the 24.3 MP digital image camera (Sony A60 0 0,) was used to acquire the raw image of the maize as well as the weeds in situ on various farmlands within the study area. The images, as well as video frames, were acquired at a distance of one foot from the crop while maintaining the camera at a uniform height.

Value of the Data
This data has immense value to the scientific community. The following points highlights some of the reasons why this data is important to the scientific community: • The published weed database will provide baseline information to researchers working on weed recognition and control, using computer vision and machine learning for intelligent herbicide application in maize for increased crop yield. • The dataset generated will be relevant to weed scientists, agronomists, botanists and farmers interested in precision and smart agriculture, involving automated weed recognition and control in maize farmland. • Government agencies, policymakers and agribusiness will also be guided on policies and programs that will drive the development, adoption and regulation of technologies and products that are likely to be developed with the availability of this data. • The dataset can also be used for weed species and maize plant recognition by annotating the area of interest, extracting relevant features and training a machine/deep learning model for recognition. • The trained model from the dataset can be used to develop autonomous systems for herbicide spraying and other related applications.

Objective
The dataset was acquired locally in Niger State, North Central Nigeria, covering four local government areas. The goal is to localize the dataset in order to meet the urgent demands of the local farmers. Due to the varying types of soil and land in those places, the authors found that different agricultural practices were implemented in the various local government districts that were visited. In addition, review of literatures in [3][4][5][6][7][8][9][10] , indicated that data of this nature were not widely available; therefore, the authors had to curate a dataset for the public and save it in a public domain for use.

Data Description
The dataset in [2] , contains images of maize plants and weed species. The dataset contains 36,874 images in total and is stored in four folders: Annotated Maize-Weed Images, Data Description and Questionnaire, Dry Season Maize Weed Images, and Wet Season Maize Weed Images. The Dry Season contains 18,187 images captured during the dry season farm survey. The Wet Season contains 18,187 images captured during the wet season farm survey, and the Annotated contains 500 annotated images selected from the Dry Season survey saved in JSON, XML, and txt format. The images of the raw wet and dry seasons were captured using a highresolution digital camera during the Maize-Weed survey carried out on 18 farm locations in the North Central part of Nigeria. In contrast, the annotation of the images was achieved using the Labelmg suite, an open-source annotation tool. The summary of the data is presented in Table 1 .

Experimental Design, Materials and Methods
Several farms in four (Bosso, Gbako, Katcha, and Lapai) different Local Government Areas (LGAs) in Niger State, Nigeria were visited. The details of the farms visited and their geographic information are shown in Table 2 . A total of 18 farms were visited with the lowest elevation of 76 metres above sea level(masl) and the highest elevation of 225 metres above sea level (masl). The longitude and latitude are also shown in the Table 2 . Primary data were collected from the respondents during the field visits (survey) with use of structured questionnaire.
Maize-based cropping system farms were surveyed in Bosso, Gbako, Katcha and Lapai LGAs ( Fig. 1 ) between February and March 2022. A handheld Global Positioning System machine (GPS-4300; Ethrex 10 Garmin, Taiwan) was used to record the coordinates of each location, and the data were used to map the surveyed locations [ 1 , 11 ].
The dataset contains images of maize crops and their accompanying weeds that alongside farm. These images were taken on field trips to 18 different maize farmlands within four (4) Local Government Areas of Niger State, Nigeria, West Africa. The data were collected from irrigated maize fields under traditional maize cultivation practices. Raw data of maize and weeds were captured in situ using a 24.3 MP digital picture camera (Sony A60 0 0). The images and video frames were taken one foot away from the crop, with the camera held at a constant height. The images were acquired in a 'handheld' manner. This implies that the photographer moved through the farmland while manually holding the camera. This manual data acquisition process was required because the terrain of the farmland and the spacing of the ridges did not allow for the use of a vehicle in which the camera could be mounted. The images were checked individually to ensure they were in focus and the blurry images removed from the dataset. Direct interviews and open discussion with the maize farmers were carried out on the status, perception of weed infestation and methods of weed control, and the need for a technology that can detect weeds and apply herbicide in their fields. This information was obtained from the maize farmers using a guided questionnaire during the survey. After data acquisition, the data were filtered to remove noise and outliers from the images. The filtered images were stored in the 'JPEG' format, the same as the acquired images to preserve data integrity. Furthermore, the data was annotated with the open-source annotation tool 'LabelImg' and saved as XML and JSON formats.

Ethics Statements
The authors collected data from various farms in the study area. The farm owners were contacted, and permission was obtained from them by the research team. In addition, the surveys were conducted in the presence of the farm owners, and they were interviewed accordingly. Sample of the administered questionnaire is available in the data repository at https: //data.mendeley.com/datasets/jjbfcckrsp .

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.