High Speed Crop and Weed Identification in Lettuce Fields for Precision Weeding

Abstract

Precision weeding can significantly reduce or even eliminate the use of herbicides in farming. To achieve high-precision, individual targeting of weeds, high-speed, low-cost plant identification is essential. Our system using the red, green, and near-infrared reflectance, combined with a size differentiation method, is used to identify crops and weeds in lettuce fields. Illumination is provided by LED arrays at 525, 650, and 850 nm, and images are captured in a single-shot using a modified RGB camera. A kinematic stereo method is utilised to compensate for parallax error in images and provide accurate location data of plants. The system was verified in field trials across three lettuce fields at varying growth stages from 0.5 to 10 km/h. In-field results showed weed and crop identification rates of 56% and 69%, respectively. Post-trial processing resulted in average weed and crop identifications of 81% and 88%, respectively.

A Deep Learning Approach for Weed Detection in Lettuce Crops Using Multispectral Images

“…Otherwise, Elstone et al [41], has her own approach of using a robotic system for detecting lettuces trough a multispectral camera and illumination control system. This approach focuses on NIR reflectivity pixels values and a size classificator assuming that crops will be always larger than weeds.…”

“…Table 4 presents a comparison between the methods that were mentioned and the approaches presented here. Another important remark about the comparison established in Table 4 is related with the use of deep learning models for weed detection instead of classical image processing techniques reflected in [40,41]. This is because deep learning models have great generalization capabilities and can be used with different kinds of lettuce crops, no matter the geographical location.…”

“…Meanwhile, those methods that include convolutional neural networks such as YOLO and Mask R-CNN can get the main features of a lettuce by themselves, even features that are outside of human visual perception, no matter how many weeds are in the picture. This advantage is very useful when the weed is the same size as the lettuce-solving in this way the limitation presented in [41]. Finally, the fact that this approach does not involve the construction of a complex system that needs lighting control and that it works on a drone allows greater flexibility with quite acceptable results regarding its performance as a weed detector.…”

“…Previous works in lettuce fields such as those of Raja et al [40] and Elstone et al Raja et al [41] obtained good results for weed and crop identification using RGB and multiespectral images, but weeds in highland tropical conditions have different forms and grow in big patches, making detection difficult; therefore, we developed methods to firstly perform vegetation detection, secondly perform crop detection, and lastly identify and quantify weeds in the field.…”

Review of Weed Detection Methods Based on Computer Vision

“…Some scholars have also used Vis–NIR to classify weeds in crops, but studies are limited to laboratory feasibility studies and rely extensively on stoichiometry to select effective wavelengths and establish calibration models [ 78 , 79 ]. Elstone et al [ 80 ] achieved good results in the identification of weeds and crops by using RGB and multispectral images in a lettuce field. However, weeds in plateau tropical conditions have different shapes and grow in large blocks, such that detecting them is difficult.…”