Vegetation Indices (VIs) obtained from remote sensing based canopies are quite simple and effective algorithms for quantitative and qualitative evaluations of vegetation cover, vigor, and growth dynamics, among other applications. These indices have been widely implemented within RS applications using different airborne and satellite platforms with recent advances using Unmanned Aerial Vehicles (UAV). Up to date, there is no unified mathematical expression that defines all VIs due to the complexity of different light spectra combinations, instrumentation, platforms, and resolutions used. Therefore, customized algorithms have been developed and tested against a variety of applications according to specific mathematical expressions that combine visible light radiation, mainly green spectra region, from vegetation, and nonvisible spectra to obtain proxy quantifications of the vegetation surface. In the real-world applications, optimization VIs are usually tailored to the specific application requirements coupled with appropriate validation tools and methodologies in the ground. The present study introduces the spectral characteristics of vegetation and summarizes the development of VIs and the advantages and disadvantages from different indices developed. This paper reviews more than 100 VIs, discussing their specific applicability and representativeness according to the vegetation of interest, environment, and implementation precision. Predictably, research, and development of VIs, which are based on hyperspectral and UAV platforms, would have a wide applicability in different areas.

Agriculture is facing severe challenges from crop stresses, threatening its sustainable development and food security. This work exploits aerial visual perception for yellow rust disease monitoring, which seamlessly integrates state-of-the-art techniques and algorithms including UAV sensing, multispectral imaging, vegetation segmentation and deep learning U-Net. A field experiment is designed by infecting winter wheat with yellow rust inoculum, on top of which multispectral aerial images are captured by DJI Matrice 100 equipped with RedEdge camera. After image calibration and stitching, multispectral orthomosaic is labelled for system evaluation by inspecting high-resolution RGB images taken by Parrot Anafi Drone. The merits of the developed framework drawing spectral-spatial information concurrently are demonstrated by showing improved performance over purely spectral based classifier by the classical random forest algorithm. Moreover, various network input band combinations are tested including three RGB bands and five selected spectral vegetation indices by Sequential Forward Selection strategy of Wrapper algorithm.

Wheat stripe rust is one of the main wheat diseases worldwide, which has significantly adverse effects on wheat yield and quality, posing serious threats on food security. Disease severity grading plays a paramount role in stripe rust disease management including breeding disease-resistant wheat varieties. Manual inspection is time-consuming, labor-intensive and prone to human errors, therefore, there is a clearly urgent need to develop more effective and efficient disease grading strategy by using automated approaches. However, the differences between wheat leaves of different levels of stripe rust infection are usually tiny and subtle, and, as a result, ordinary deep learning networks fail to achieve satisfying performance. By formulating this challenge as a fine-grained image classification problem, this study proposes a novel deep learning network C-DenseNet which embeds Convolutional Block Attention Module (CBAM) in the densely connected convolutional network (DenseNet). The performance of C-DenseNet and its variants is demonstrated via a newly collected wheat stripe rust grading dataset (WSRgrading dataset) at Northwest A&F University, Shaanxi Province, China, which contains a total of 5,242 wheat leaf images with 6 levels of stripe rust infection. The dataset was collected by using various mobile devices in the natural field condition. Comparative experiments show that C-DenseNet with a test accuracy of 97.99% outperforms the classical DenseNet (92.53%) and ResNet (73.43%). GradCAM++ network visualization also shows that C-DenseNet is able to pay more attention to the key areas in making the decision. It is concluded that C-DenseNet with an attention mechanism is suitable for wheat stripe rust disease grading in field conditions.

High throughput phenotyping (HTP) for wheat (Triticum aestivum L.) stay green (SG) is expected in field breeding as SG is a beneficial phenotype for wheat high yield and environment adaptability. The RGB and multispectral imaging based on the unmanned aerial vehicle (UAV) are widely popular multi-purpose HTP platforms for crops in the field. The purpose of this study was to compare the potential of UAV RGB and multispectral images (MSI) in SG phenotyping of diversified wheat germplasm. The multi-temporal images of 450 samples (406 wheat genotypes) were obtained and the color indices (CIs) from RGB and MSI and spectral indices (SIs) from MSI were extracted, respectively. The four indices (CIs in RGB, CIs in MSI, SIs in MSI, and CIs + SIs in MSI) were used to detect four SG stages, respectively, by machine learning classifiers. Then, all indices’ dynamics were analyzed and the indices that varied monotonously and significantly were chosen to calculate wheat temporal stay green rates (SGR) to quantify the SG in diverse genotypes. The correlations between indices’ SGR and wheat yield were assessed and the dynamics of some indices’ SGR with different yield correlations were tracked in three visual observed SG grades samples. In SG stage detection, classifiers best average accuracy reached 93.20–98.60% and 93.80–98.80% in train and test set, respectively, and the SIs containing red edge or near-infrared band were more effective than the CIs calculated only by visible bands. Indices’ temporal SGR could quantify SG changes on a population level, but showed some differences in the correlation with yield and in tracking visual SG grades samples. In SIs, the SGR of Normalized Difference Red-edge Index (NDRE), Red-edge Chlorophyll Index (CIRE), and Normalized Difference Vegetation Index (NDVI) in MSI showed high correlations with yield and could track visual SG grades at an earlier stage of grain filling. In CIs, the SGR of Normalized Green Red Difference Index (NGRDI), the Green Leaf Index (GLI) in RGB and MSI showed low correlations with yield and could only track visual SG grades at late grain filling stage and that of Norm Red (NormR) in RGB images failed to track visual SG grades. This study preliminarily confirms the MSI is more available and reliable than RGB in phenotyping for wheat SG. The index-based SGR in this study could act as HTP reference solutions for SG in diversified wheat genotypes.

Accurately detecting and segmenting grape cluster in the field is fundamental for precision viticulture. In this paper, a new backbone network, ResNet50-FPN-ED, was proposed to improve Mask R-CNN instance segmentation so that the detection and segmentation performance can be improved under complex environments, cluster shape variations, leaf shading, trunk occlusion, and grapes overlapping. An Efficient Channel Attention (ECA) mechanism was first introduced in the backbone network to correct the extracted features for better grape cluster detection. To obtain detailed feature map information, Dense Upsampling Convolution (DUC) was used in feature pyramid fusion to improve model segmentation accuracy. Moreover, model generalization performance was also improved by training the model on two different datasets. The developed algorithm was validated on a large dataset with 682 annotated images, where the experimental results indicate that the model achieves an Average Precision (AP) of 60.1% on object detection and 59.5% on instance segmentation. Particularly, on object detection task, the AP improved by 1.4% and 1.8% over the original Mask R-CNN (ResNet50-FPN) and Faster R-CNN (ResNet50-FPN). For the instance segmentation, the AP improved by 1.6% and 2.2% over the original Mask R-CNN and SOLOv2. When tested on different datasets, the improved model had high detection and segmentation accuracy and inter-varietal generalization performance in complex growth environments, which is able to provide technical support for intelligent vineyard management.

Disease and pest detection of grape foliage is essential for grape yield and quality. RGB image (RGBI), multispectral image (MSI), and thermal infrared image (TIRI) are widely used in the health detection of plants. In this study, we collected three types of grape foliage images with six common classes (anthracnose, downy mildew, leafhopper, mites, viral disease, and healthy) in the field. ShuffleNet V2 was used to build up detection models. According to the accuracy of RGBI, MSI, TIRI, and multi-source data concatenation (MDC) models, and a multi-source data fusion (MDF) decision-making method was proposed for improving the detection performance for grape foliage, aiming to enhance the decision-making for RGBI of grape foliage by fusing the MSI and TIRI. The results showed that 40% of the incorrect detection outputs were rectified using the MDF decision-making method. The overall accuracy of MDF model was 96.05%, which had improvements of 2.64%, 13.65%, and 27.79%, compared with the RGBI, MSI, and TIRI models using label smoothing, respectively. In addition, the MDF model was based on the lightweight network with 3.785 M total parameters and 0.362 G multiply-accumulate operations, which could be highly portable and easy to be applied.