Wheat-Net: An Automatic Dense Wheat Spike Segmentation Method Based on an Optimized Hybrid Task Cascade Model.

Brian J. Steffenson,Wen-Hao Su,Jian Wei,James Anderson,Cory D. Hirsch,Ce Yang,Qin Ma,Jiajing Zhang,An Min

doi:10.3389/fpls.2022.834938

Abstract

Precise segmentation of wheat spikes from a complex background is necessary for obtaining image-based phenotypic information of wheat traits such as yield estimation and spike morphology. A new instance segmentation method based on a Hybrid Task Cascade model was proposed to solve the wheat spike detection problem with improved detection results. In this study, wheat images were collected from fields where the environment varied both spatially and temporally. Res2Net50 was adopted as a backbone network, combined with multi-scale training, deformable convolutional networks, and Generic ROI Extractor for rich feature learning. The proposed methods were trained and validated, and the average precision (AP) obtained for the bounding box and mask was 0.904 and 0.907, respectively, and the accuracy for wheat spike counting was 99.29%. Comprehensive empirical analyses revealed that our method (Wheat-Net) performed well on challenging field-based datasets with mixed qualities, particularly those with various backgrounds and wheat spike adjacence/occlusion. These results provide evidence for dense wheat spike detection capabilities with masking, which is useful for not only wheat yield estimation but also spike morphology assessments.

Highlights

Wheat is the most widely cultivated cereal crop and one of the most important food sources for humans in the world
It should be noted that the main goal of this paper is to accurately segment wheat spikes in complex environments, so the datasets and scenarios may different from pure wheat counting studies
MultiStepLR was better than CosineAnnealingLR in terms of accuracy and speed for our model of wheat spike detection, so we chose MultiStepLR to decay the learning rate and further improve the performance of our model

Summary

Introduction

Wheat is the most widely cultivated cereal crop and one of the most important food sources for humans in the world. The spike is the most important component of the wheat plant because it contains the seeds that are harvested and consumed. In-field automated wheat spike detection based on remote sensing is an important step toward yield estimation and spike morphology assessments. The remote sensing imaging devices are useful tools to replace traditional artificial detection (Aparicio et al, 2000). The cheaper RGB imaging camera is a realistic alternative to achieve effective wheat detection. Deep learning (DL) with strong feature learning abilities has spawned a multitude of applications in RGB images. It encodes the composition of lower-level features into more discriminative higher-level features (Ni et al, 2019). DL, with its advantages of high precision and intelligence, is an attractive alternative to conventional wheat spike detection methods (Germain et al, 1995; Cointault et al, 2008a,b)

Objectives

Methods

Results

Discussion

Conclusion