Development of an intelligent laser biospeckle system for early detection and classification of soybean seeds infected with seed-borne fungal pathogen (Colletotrichum truncatum). (December 2021)

• Record Type: Journal Article
• Authors: Singh, Puneet; Chatterjee, Amit; Rajput, Laxman S.; Rana, Santosh; Kumar, Sanjeev; Nataraj, Vennampally; Bhatia, Vimal; Prakash, Shashi
• Is Part Of: Biosystems engineering. Volume 212(2021)
• Page Start: 442
• Page End: 457
• Publication Date: 2021-12
• Subjects: Agriculture -- Biospeckle Analysis -- Early Detection -- Laser -- Machine Learning -- Seed-Borne Fungi; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2021.11.002
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : There is a need for developing rapid and non-destructive techniques for the early detection of seed-borne fungal pathogen because they can be an essential step towards adopting effective disease control measures. Existing techniques for detecting seed-borne diseases have poor sensitivity towards early stages of pathogen development (i.e., when seeds are asymptomatic) and they are also expensive, time-consuming, complex, require mycological skills and destructive testing operations. Aiming at overcoming the above limitations of the existing techniques, a novel laser biospeckle based method is proposed for early detection of seed-borne fungal infection in conjunction with machine learning. Soybean seeds infected by low concentrations (10 2 -10 6 spores ml −1 ) of Colletotrichum truncatum were analysed by using full field biospeckle analysis to establish the possible relationship between biological activity in early stages of pathogen infection, with and without the use of frequency filtering. The results demonstrate that the biospeckle activity (BA), for both, raw and frequency filtered data was significantly high (p < 0.05) for the diseased seeds even for low inoculum concentrations. Moreover, the amplitude values of mid frequency spectral components for diseased seeds were higher than those of lower and higher spectral components which correspond to the BA of fungal infected seeds. Several classical machine learning algorithms were trained to model the response of healthy and diseased samples after parameter optimisation. Obtained results showed that k-nearest neighbour (k-NN), decision tree (DT), and artificial neural network (ANN) based predictive models presented strong robustness and high performance with overall accuracy reaching up to 96.94% for classifying diseased seeds. Highlights: Early identification of seed-infection is performed by using biospeckle technique. Frequency decomposition of data is performed to isolate fungal activity. Machine learning based automatic classification of infected seeds is performed. High classification accuracy (96.94%) is obtained for classifying the seeds. The proposed method is non-destructive, fast, automatic, and full field.