Images, features, or feature distributions? A comparison of inputs for training convolutional neural networks to classify lentil and field pea milling fractions

McDonald, Linda S.; Assadzadeh, Sahand; Panozzo, JF

doi:10.1016/j.biosystemseng.2021.05.011

Cited by 10 publications

(6 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Grain quality when received by grain receival agents is graded using two approaches: firstly, subjective procedures undertaken by trained operators (grain inspectors) assessing visual traits, including stained, cracked, defective grain and contaminants; and secondly objectively using standardised instrumentation, such as sieves to determine grain size and near infrared spectroscopy (NIR) to determine composition, such as protein and moisture concentration. Key traits used in valuing and trading grain include the percentage of small grain (screenings), foreign material (unable to be processed, milled or malted), contaminates, sprouted, stained or discoloured grain, broken, damaged or distorted grain, presence of insects or mould, test weight, or composition including moisture concentration, protein concentration, low levels of aflatoxins and a high Hagberg falling number test [8][9][10][11]. Many of these traits are measured as a % per weight of samples subsampled from the grain load.…”

Section: Of 24mentioning

confidence: 99%

Technologies and Data Analytics to Manage Grain Quality On-Farm: A Review

Walker¹,

Assadzadeh²,

Wallace³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

Section: Of 24mentioning

confidence: 99%

Technologies and Data Analytics to Manage Grain Quality On-Farm: A Review

Walker¹,

Assadzadeh²,

Wallace³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The predicted dehulling efficiencies were highly correlated (R 2 = 0.9 and RMSE < 2%) with the laboratory method. Additionally, a DIA approach to the assessment of split and dehulled grain yields of milled lentil and field pea products was presented by McDonald et al (2021), with an overall validation classification accuracy of 88.1%.…”

Section: For Pulse Evaluationmentioning

confidence: 99%

“…The destructive and time‐consuming nature of manual milling‐quality assessments, often referred to as dehulling the seed and splitting the cotyledon, makes MV approaches an appealing option within the breeding selection process. Typically, fewer than 15 × 100 g samples of dehulled lentil and split cotyledon can be hand‐sorted per day, so testing efficiency for split yields would increase substantially through MV assessments (McDonald et al, 2021). Predictions of milling potential from whole‐grain images would enable assessments of processing quality on early generation germplasm when the volume of grain is insufficient for destructive testing.…”

Section: For Pulse Evaluationmentioning

confidence: 99%

A review of the opportunities for spectral‐based technologies in post‐harvest testing of pulse grains

McDonald

Panozzo

2022

Legume Science

Self Cite

View full text Add to dashboard Cite

Pulse grains are phenotypically diverse varying in colour, size, shape, and uniformity and have been integrated within many cultures and cuisines for several thousand years. Consumption of pulses within traditional dishes is still the dominant use for these grains, and therefore, the marketability is largely based on visual characteristics. There is also increasing interest into the utilisation of pulses in new processed food products because of their high protein content.Pulse‐quality assessment is critical within industry to determine marketability of the produce and remuneration for growers; however, the methods for assessment are largely subjective, completed by visual appraisal. Furthermore, targeted pulse‐quality traits form part of the overall strategy of plant breeding programmes, but the grain‐assessment methodologies are time consuming, constraining testing efficiency, and some destructive tests are reserved for advanced germplasm.Recent advances in computing and spectral sensing technology have improved opportunities for development of non‐destructive, high‐throughput and accurate machine vision (MV) systems for product‐quality evaluation. Algorithms based on digital image analysis have been developed to classify and quantify characteristics relating to the size, shape, colour and defects of grains and other agricultural products. Additionally, near‐infrared‐spectral processing has been successfully applied in the prediction of compositional constituents, such as protein and moisture, for some agricultural products.This review describes the standard methodologies for the assessment of pulse‐quality traits and developments in MV applications for grain quality assessment. Opportunities are identified, both within the pulse grain industry and plant breeding programmes, for objective and standardised post‐harvest testing of pulse grains through MV.

show abstract

“…Furthermore, they can achieve high performances for different classification and detection problems, achieving faster inference time and higher detection rates than traditional computer vision methods 12 . Thus, the association of images to convolutional neural networks has already been used for the automatic quantification of ears of wheat in the field 13 and for the classification of the milling fraction of lentils and peas 14 .…”

Section: Introductionmentioning

confidence: 99%

Convolutional neural networks in the qualitative improvement of sweet potato roots

Fernandes

Valadares

Rodrigues

et al. 2023

Sci Rep

View full text Add to dashboard Cite

The objective was to verify whether convolutional neural networks can help sweet potato phenotyping for qualitative traits. We evaluated 16 families of sweet potato half-sibs in a randomized block design with four replications. We obtained the images at the plant level and used the ExpImage package of the R software to reduce the resolution and individualize one root per image. We grouped them according to their classifications regarding shape, peel color, and damage caused by insects. 600 roots of each class were destined for training the networks, while the rest was used to verify the quality of the fit. We used the python language on the Google Colab platform and the Keras library, considering the VGG-16, Inception-v3, ResNet-50, InceptionResNetV2, and EfficientNetB3 architectures. The InceptionResNetV2 architecture stood out with high accuracy in classifying individuals according to shape, insect damage, and peel color. Image analysis associated with deep learning may help develop applications used by rural producers and improve sweet potatoes, reducing subjectivity, labor, time, and financial resources in phenotyping.

show abstract

Images, features, or feature distributions? A comparison of inputs for training convolutional neural networks to classify lentil and field pea milling fractions

Cited by 10 publications

References 30 publications

Technologies and Data Analytics to Manage Grain Quality On-Farm: A Review

Technologies and Data Analytics to Manage Grain Quality On-Farm: A Review

A review of the opportunities for spectral‐based technologies in post‐harvest testing of pulse grains

Convolutional neural networks in the qualitative improvement of sweet potato roots

Contact Info

Product

Resources

About