Computer Vision Methods in Precision Agriculture
1South China Agriculture University, Guangzhou, China
2The University of Tennessee, Knoxville, USA
3Northwest A&F University, Yangling, China
4Zhongkai University of Agriculture and Engineering, Guangzhou, China
5Foshan University, Foshan, China
Computer Vision Methods in Precision Agriculture
Description
Computer vision is a technological application that can detect, locate, or track objects. It has been extensively studied in industrial and precision agriculture fields, particularly regarding autonomous driving, surface defect detection, object detection and localization, automatic harvesting, robotics, plant phenotyping, and crop yield estimation. Autonomous driving allows a motor vehicle to sense its environment and operate without human involvement. Uncertainties on the road such as illumination changes and occlusion make autonomous driving highly challenging. For example, pedestrian detection and lane line detection are difficult to achieve. Surface defect detection can distinguish desired features from anomalies. It is an important technology in terms of production automation. Surface defect detection requires an algorithm to detect and analyze defects quickly and robustly, which is highly challenging. Deep learning techniques can facilitate object detection and localization, but deep learning algorithms require burdensome computational processes and large amounts of storage, which poses a significant challenge in their implementation.
Mobile and fast deep learning algorithms are preferred. Automatic fruit/vegetable harvesting robots have been researched for several decades, but there is still no such commercial product available. Existing harvesting robots struggle with target fruit/vegetable detection and localization, obstacle detection and localization, collision-free path planning, vision-based servo control, and other problems. The “plant phenotype” refers to all measurable features of a plant (e.g., leaf colour and shape). The phenotype describes the relationship between the genotype and the environment of a plant’s measurable characteristics. Measuring plant phenotypes via computer vision is a notable potential research direction as it may allow for the accurate detection and segmentation of plant parts as well as the accurate reconstruction of plant parts. In addition, because nonlinear filtering is indispensable for achieving object localization and tracking, object detection under camera occlusion, as well as for fusing images from distributed cameras, nonlinear filtering methods (such as nonlinear Kalman filter, particle filter, and distributed filters) are also preferred.
The aim of this Special Issue is to bring together original research articles and review articles highlighting the application of computer vision in precision agriculture. Submissions focusing on autonomous driving, surface defect detection, object detection and localization, automatic harvesting, robotics, plant phenotyping, and crop yield estimation are particularly welcome.
Potential topics include but are not limited to the following:
- Computer vision-based autonomous driving algorithms, which can robustly detect pedestrians, lane lines, and other targets
- Application of generalizable real-time surface defect detection methods in precision agriculture
- Application of end-to-end, real-time, deep neural networks that can detect and locate objects simultaneously in precision agriculture
- Fruit/vegetable detection and localization for automatic harvesting robots (e.g., end-to-end, real-time, and precise deep neural networks that can detect target produce in a field)
- Generalizable methods for segmenting or reconstructing plant parts to support plant phenotyping processes
- Crop yield estimation using computer vision technology
- Advanced nonlinear filtering methods in precision agriculture