Development of sugarcane and trash identification system in sugar production using hyperspectral imaging

Abstract

Classification and differentiation of clean sugarcane from trash (green sugarcane leaf, dry sugarcane leaf, stone, and soil) are important for the sugar payment system at a sugar mill. Currently, the methods used to do this are manual and subjective. Therefore, this study is aimed at accurately differentiating clean sugarcane from trash by using hyperspectral imaging with multivariate analyses. Samples containing sugarcane billets and trash mixed in a ratio of 18:38 were analyzed in this study. The reflectance data of the samples were analyzed in the wavelength range of 400–1000 nm via principal component analysis (PCA). The PCA model was capable of identifying all of the clean sugarcane and trash samples. The spectral loadings of the PCA model show that the sugarcane and trash samples are easily identifiable based on the color (visible light) of each class, water absorption (approximately 970 nm), and chlorophyll absorption (approximately 680 nm). Based on the characteristic wavelengths of the PCA loading peaks, over 90% of the sugarcane and trash samples were differentiated using a multiple linear regression model. Sugarcane and trash are classified by using partial least-squares discriminant analysis and support vector machine models. For all wavelengths, the classification rate is 92.9% and 98.2%, respectively. This shows that sugarcane and trash can be accurately classified and differentiated by using hyperspectral imaging and multivariate analyses.

Detection of breakage and impurity ratios for raw sugarcane based on estimation model and MDSC-DeepLabv3+

“…The experimental results demonstrated correlation coefficients of 0.98, 0.93, and 0.91 for the training, validation, and test sets, respectively. Aparatana et al (2020) employed principal component analysis (PCA), PLS-DA, and support vector machine (SVM) to classify and differentiate sugarcane and impurities, including green leaf, dry leaf, stone, and soil, based on their spectral information. The research findings indicated that PCA, PLS-DA, and SVM achieved classification rates of 90%, 92.9%, and 98.2%, respectively.…”

“…From the perspective of recognition tasks, the aforementioned studies can be categorized into three types: image classification, object detection, and semantic segmentation. Image classificationbased approaches (Momin et al, 2017;Guedes and Pereira, 2019;Shen et al, 2019;Aparatana et al, 2020;Chen et al, 2020;Guedes et al, 2020;Dos Santos et al, 2021; cannot capture pixel-level information for subsequent construction of a mass-pixel fitting model. Object detection can be utilized for real-time classification and localization of crops and impurities (Zhang et al, 2022;Xu et al, 2023;, but they still cannot support subsequent mass estimation based on pixels of detected objects.…”

Non-destructive Laboratory Analysis of the Detection of Unhealthy Sugarcane Stalks Using a Portable Vis-NIR Spectrometer

“…Spectroscopic techniques are widely used in the food and agricultural industries for detection and identification owing to them being fast, accurate, and cost-efficient. The spectrum can also be used to identify sugarcane or predict suger content (Aparatana et al, 2020;Maraphum et al, 2020). A portable visible near-infrared (Vis-NIR) spectrometer has been recently developed and was used to analyze sugarcane because it is reproducible, non-destructive, and convenient for field research (Maraphum et al, 2018;Phuphaphud et al, 2020;Phuphaphud et al, 2019;Sanseechan et al, 2018;Taira et al, 2013;Taira et al, 2015).…”