Elsevier

Biosystems Engineering

Volume 188, December 2019, Pages 165-177
Biosystems Engineering

Research Paper
Peanut maturity classification using hyperspectral imagery

https://doi.org/10.1016/j.biosystemseng.2019.10.019Get rights and content

Highlights

  • Peanut maturity estimated without pod blasting or a maturity profile board.

  • Hyperspectral imagery is used to estimate pixel-level maturity.

  • Trained classification model is applied to peanuts of different years and cultivars.

  • Peanut maturity classification on testing data averages of 91.15% overall accuracy.

Seed maturity in peanut (Arachis hypogaea L.) determines economic return to a producer because of its impact on seed weight (yield), and critically influences seed vigour and other quality characteristics. During seed development, the inner mesocarp layer of the pericarp (hull) transitions in colour from white to black as the seed matures. The maturity assessment process involves the removal of the exocarp of the hull and visually categorizing the mesocarp colours into varying colour classes from immature (white, yellow, orange) to mature (brown, and black). This visual colour classification is time consuming because the exocarp must be manually removed. In addition, the visual classification process involves human assessment of colours, which leads to large variability of colour classification from observer to observer. A more objective, digital imaging approach to peanut maturity is needed, optimally without the requirement of removal of the hull's exocarp. This study examined the use of a hyperspectral imaging (HSI) process to determine pod maturity with intact pericarps. The HSI method leveraged spectral differences between mature and immature pods within a classification algorithm to identify the mature and immature pods. Therefore, there is no need to remove the exocarp nor is there a need for subjective colour assessment in the proposed process. The results showed a consistent high classification accuracy using samples from different years and cultivars. In addition, the proposed method was capable of estimating a continuous-valued, pixel-level maturity value for individual peanut pods, allowing for a valuable tool that can be utilized in seed quality research. This new method solves issues of labour intensity and subjective error that all current methods of peanut maturity determination have.

Introduction

Cultivated peanut (Arachis hypogaea L.) is an important agronomic legume grown mainly in tropical and subtropical areas. Peanut agronomic production can be complex due to the crop's unique geocarpic reproduction, susceptibility to many fungal pathogens, and complex mechanical harvest procedure requiring separate inversion and threshing operations. Further, peanut has an indeterminate growth habit, thus having pods at various maturity levels throughout a growing season and at harvest time. It is critical for growers to accurately determine pod maturity to be able to maximize seed weight and quality, thus leading to optimal economic returns. However, the geocarpic fruit habit of peanut increases the difficulty in determining the percent of pods that are mature and ultimately to determining the optimum digging time. If digging occurs too early in development, immature pods can cause poor yield, grade, seed quality and flavour. If digging occurs too late, over-mature pods can detach from the vine during the digging or the threshing process.

Several methods have been developed for growers to predict and optimize harvest time. The most commonly accepted method is to remove the exocarp from the pericarp (hull) and categorize the inner mesocarp colour. This method was standardized by Williams and Drexler (1981) who created a Maturity Profile Board (MPB) that provided colour classification into five main colour categories (white, yellow, orange, brown, and black) and sub-categories within representing various shades within the main colour categories. Mesocarp colours of black and brown indicate mature seeds and conversely, orange, yellow, and white represent immature seeds. Based on the ratio of pods within various main and shade categories, growers can determine the optimal number of days until digging the crop (the first stage of harvest). This remains the primary method utilized by producers and researchers to date for evaluating seed maturity.

However, the MPB method has many flaws and disadvantages. First, it involves exocarp removal, often requiring “pod blasting” using a pressure washer. In this process, most white pods are blown apart and lost in the analysis due to their fragile pericarp structure and high water content. Further, this process can be extremely time consuming to reach the level of exocarp removal for accurate visual mesocarp assessment while not destroying too many of the pods in the process. Second, once the exocarp is removed, pods must be colour categorized using the human eye, a process that is extremely subjective (due to observer variability in colour categorization, lighting conditions, observer fatigue, etc.), thus introducing large error potential into the process. In addition, the process of visually categorizing pods can be very time consuming as well because it involves the individual placement of 100–150 pods within main and shade colour categories. For these reasons, the current process for determining peanut maturity is flawed and in need of a method that is objective and does not require exocarp removal.

A peanut pod maturity estimation approach was recently proposed (Bindlish, Abbott, & Balota, 2017) to replace the subjective colour categorization process with a simple Nearest Neighbour classifier that classifies the median RGB colour of mesocarp of each peanut into one of ten pre-set colour classes (i.e. ten maturity levels). In other words, it learns a simple mapping from RGB values of mesocarp images to peanut maturity. However, the time-consuming pod blasting step to remove the exocarp is still necessary in this method.

This challenge, to estimate peanut maturity on the exocarp level without pod blasting, lends itself to an evaluation of the peanut pericarp using more sophisticated imaging techniques. The current project explored using a hyperspectral imaging (HSI) process capable of detecting materials that differed in chemical composition. The colour change process in peanut involves the accumulation of tannins and other polyphenols within the mesocarp layer. It was reported that in hulls and seed coats, the tannin content increased significantly as peanut pod developed and showed a close relationship between tannin and maturity (Sanders, 1977), thus leading to a change in chemical composition within the pericarp that corresponds to seed development. If an HSI process could detect pericarp hyperspectral signatures indicative of each of the major colour classes, then peanut maturity could be determined non-destructively.

Section snippets

Theoretical framework of the HSI

The development of the HSI process involved the collection of a high dimensional hyperspectral image data cube. This cube consisted of a stack of hundreds of two-dimensional images collected at different wavelengths across the electromagnetic spectrum, as illustrated in Fig. 1. The spectral signature collected across all measured wavelengths associated with each pixel in a hyperspectral image is composed of the radiance values from each of the measured wavelengths and characterizes the chemical

Results and discussion

In this study, the maturity estimation and classification experiments were trained on peanuts collected in 2016 and tested on peanuts collected in 2017, and vice versa.

Conclusion

This study successfully developed and tested a peanut maturity classification system that involved an objective, non-destructive technique utilising HSI. In this method, we were able to evaluate individual pod maturity with intact exocarp layers, thus eliminating the need for peanut blasting. We found that peanuts at different maturity levels have different spectral signatures when imaged with a hyperspectral sensor. Therefore, we were able to leverage spectrum difference to estimate pod

Declaration of competing interest

None declared.

Acknowledgement

This work was supported by National Peanut Board, USA; Florida Peanut Producers Association, and National Science Foundation, Division of Information and Intelligent Systems, USA [grant number #1723891 “CAREER: Supervised Learning for Incomplete and Uncertain Data”].

Nomenclature table

ci
Confidence value of ith image region, a scalar in R+
D
The dimensionality of pixel spectrum, an integer in R+
ek
k-th endmember spectrum, a column vector in RD
E
A matrix of M endmembers, E={e1,e2,,eM}, a matrix in RD×M
εi
Noise term fori-th pixel spectrum, a vector in RD
M
The number of endmembers, an integer in R+
N
The number of pixels, an integer in R+
pik
The proportion of ek ini-th pixel spectrum, a scalar in R of [0,1]
P
A collection of pik, a matrix in RM×N
τ
Classification threshold, a scalar in R of

References (26)

  • D.B. Egli et al.

    Species differences in seed water status during seed maturation and germination

    Seed Science Research

    (1997)
  • S.A. El Rahman

    Hyperspectral image classification using unsupervised algorithms

    International Journal of Advanced Computer Science and Applications

    (2016)
  • R. Gebbers et al.

    Precision agriculture and food security

    Science

    (2010)
  • Cited by (25)

    View all citing articles on Scopus
    1

    These authors contributed equally to this work.

    View full text