Nondestructive Evaluation of Agro-products by Intelligent Sensing Techniques

Property, quality and safety assessment of agro-products are increasingly gaining attention due to the potential human health concern as well as social sustainable development. Emerging techniques and methods have particular advantages in nondestructive evaluation of agro-products due to their simplicity and faster response time, and reliable results, compared with the conventional visual inspection and destructive methods. This chapter briefly elaborates the principles and system components of some representative techniques, in particular, near infrared spectroscopy, infrared spectroscopy, fluorescence spectroscopy, Raman spectroscopy, laser induced breakdown spectroscopy, traditional machine vision, hyperspectral and multispectral imaging, magnetic resonance imaging, X-ray imaging, thermal imaging, light backscattering imaging, electrical nose and acoustics. The recent applications and technical challenges for these representative techniques are also presented.

The quality of agro-products is the foremost current issue for the food industry and consumers. Healthful agro-products such as fruits and vegetables, meat, grains, and dairy products are essential for human life, and reliable quality evaluation is important for product safety and consumer appeal. As a result, rapid and precise evaluation methods for the quality of agro-products are required. In this regard, optical sensing techniques such as imaging and spectroscopy are among the most promising techniques currently investigated for quality assessment purposes in agricultural fields. This chapter aims to present the basic concepts, components and principles of imaging and spectroscopy techniques in a comparative manner for agriculture application. Moreover, this chapter also elaborates upon the partiality of the optical sensing techniques by highlighting previous studies in agricultural applications. The insights in this chapter will help a novice to understand and encourage further knowledge about optical sensing techniques.

The quality and safety of agro-products are a global concern due to their significant role in human health and economy, and the detection of hazards or ingredients in agro-products is thus essential to ensure safety. Biosensor, as a newlyemerging but promising detection tool, has contributed a lot in this field. On the one hand, based on the high sensitivity and specificity of bio-receptors for target capture and the diversity of transducers for signal transduction, biosensors exhibit capabilities for highly sensitive, specific, accurate and rapid detection. On the other hand, the combination/integration with miniaturized and portable platforms/devices endows biosensors with unrivaled advantages in low-cost, in-field and nondestructive detection. This chapter gives a systematical introduction of biosensors for the evaluation of quality and safety of agro-products, emphasizing on new biosensing principles and the advantages of exceptional analytical performance for rapid and in-field evaluation. Recent advances in biosensors for the detection of pesticide residues, antibiotic residues, pathogenic bacteria and mycotoxins, heavy metal ions, food allergens, and ingredients in agro-products are surveyed (mainly in 2018-2020).

The internal quality of agricultural products is an important attribute that is considered by consumers when buying them. Grading agricultural products according to their internal quality, is an effective way to make the best use of the products, and thus improve the overall value. In recent years, several nondestructive, intelligent sensing techniques have been studied extensively for detecting the internal quality of agricultural products, including Vis/NIR spectroscopy, multi-/hyper-spectral imaging, nuclear magnetic resonance and imaging, X-ray and computed tomography, electrical nose and acoustic technique. In this chapter, the working principle of each technique is provided, and corresponding applications in the agricultural domain are reviewed to provide overall understanding of these techniques. The challenges and perspectives of these techniques are also analyzed.

Imaging techniques can be used to evaluate the quality and safety of agricultural products. Fusarium head blight (FHB) results in reduced barley yields and also diminished value of harvested barley. Deoxynivalenol (DON) is a mycotoxin produced by the causal Fusarium species that pose health risks to humans and livestock. DON has currently measured via gas chromatography (GC) methods that are time-consuming and expensive. We seek to apply imaging technology to rapidly and non-destructively quantify DON in high throughput and less expensive method. The feasibility of hyperspectral imaging to determine DON contents of barley kernels was evaluated using machine learning algorithms. Partial least square discriminant analysis (PLSDA) was able to discriminate kernels into four separate classes corresponding to their DON levels. Barley kernels could be classified as having low (<5 ppm) or high DON levels, with Matthews's correlation coefficient in cross-validation (M-RCV) of as high as 0.823. PLSR showed good performance in linear algorithms for DON detection, but higher accuracy was obtained by non-linear algorithms, including weighted partial least squares regression (LWPLSR), support vector machine regression (SVMR), and artificial neural network (ANN). Among all algorithms, the non-linear LWPLSR achieved the highest accuracy, with the coefficient of determination in prediction (R2 P) of 0.728 and root mean square error of prediction (RMSEP) of 3.802. The results demonstrate that hyperspectral imaging and machine learning algorithms have the potential to assist the FHB resistance breeding process by accelerating the quantification of DON in barley samples.

Optical-based technologies offer significant advantages compared with conventional methods for detecting mycotoxin and fungal contamination in agricultural and food commodities, such as rapidness and non-destructiveness. Hyperspectral imaging (HSI) integrates traditional imaging and spectroscopy technologies and thus makes it possible for high-throughput screening analysis in an onsite or on-line manner. Currently, HSI, in tandem with modern chemometrics, has demonstrated interesting and promising results for the detection of mycotoxin and fungal contamination in varieties of agricultural products. Therefore, the objective of this chapter is to give an overview of current research advances of HSI in both fluorescence and reflectance modes for the evaluation of mycotoxin and fungal contamination in agricultural and food commodities. Advances of HSI in evaluation of the main mycotoxins, including aflatoxins, ochratoxins, deoxynivalenol, fumonisins and their related fungal contaminants, are reviewed, and the results obtained from different studies are compared and discussed. Perspectives on its future trends and challenges concerning mycotoxin and fungal evaluation are also addressed.

Intelligent sensing technology of agricultural products can effectively guarantee food quality and safety, and is the key technical support to promote the rapid development of the world’s agricultural products processing industry, bringing more opportunities and development space to the emerging agricultural products processing industry. Intelligent sensing technology for agricultural products is a multidisciplinary research field, which has the advantages of fast detection speed, convenient operation, and easy online detection. This work reviews the research of optical, acoustic, electrical, magnetic, and bionic sensing technologies in the processing of agricultural products, expounds the principle, structure, and typical applications of each sensing technology, and summarizes the problems and trends in the development of each sensing technology. Intelligent sensing technology for agricultural product quality and safety is developing towards the direction of high sensitivity, automation, networking, intelligence, and multi-function, and has gradually become an indispensable and important technical means for agricultural product quality and safety inspection. The intelligent sensing technology of agricultural products is developing synchronously with the integration of the Internet of things, big data, and cloud computing, which can realize the standardization, refinement, and intelligent management of the agricultural products processing process.

Automated sorting and quality grading of agricultural produce are crucial for providing commodities with consistent quality to the consumers and markets. Machine vision has been playing a key role in this quest by presenting technological solutions that provide robust, consistent, and accurate decisions with minimal human intervention. An end-to-end quality inspection system should recognize the type of agricultural product and then perform quality grading. Accordingly, in this proof-o- -concept study, a deep learning-based end-to-end solution for quality inspection of agricultural produce is presented, where an initial system automatically sorts fruitsvegetables, while a second system grades apples by skin quality. Experimental evaluations show that the presented end-to-end solution achieves accurate and promising results, and thus holds high-potential for offering high-impact, traceable and generalizable answers for the industry.

Harvesting is one of the most challenging tasks in fruit production. Robotic fruit harvesting technologies are being studied because of labor-intensive and costly handpicking. Due to the unstructured and dynamic characteristics of both the target fruit and its surrounding environment, current harvesting robots have limited performance. Therefore, the commercial applications of most fruit harvesting robots are unrealized. The application and research progress of fruit harvesting robots in apple and kiwifruit harvesting have been reported in this chapter. The applications and development of fruit detection and end-effector design for complex orchard are focused. The main methods used in fruit detection are reviewed, including single feature detection methods, multi-features fusion detection methods, deep learning methods, and 3D reconstruction methods. The technology of end-effector design for selective harvesting with apple and kiwifruit, and shake-and-catch mechanism for bulk harvesting with apple are also reviewed. Existing research problems of fruit harvesting robots in robotic harvesting applications are mentioned, and future development directions of agriculture robots are described.

Lodging is a critical issue in wheat production, resulting in reduced yield, low crop quality, and increased difficulties in the harvest. Wheat lodging detection contributes greatly to crop management and yield estimation, as well as insurance claim issues. The current manual measurement is labor-intensive, inefficient, and subjective. Aiming to develop a more efficient and objective method to distinguish lodging from non-lodging areas, this study collected aerial color and multi-spectral images using drones attached to different cameras. The experimental field consisted of 372 wheat plots of three different sizes and three days’ datasets were collected. Individual images were first stitched to obtain an orthomosaic map and then each plot was visually classified as lodging or non-lodging. Features (i.e., color, texture, NDVI, and height) of each plot were extracted. For each day’s dataset, 300 plots (~80% of the total plots) were randomly selected to train the Support Vector Machine (SVM) model, while the remaining 72 plots (~20% of the total plots) were used to test the trained model. After training and testing 10 times, the prediction accuracy was obtained by averaging 10 prediction accuracies. When only using one feature to train the model, prediction accuracies ranged from 66% to 86%. The accuracy increased with more features incorporated for model training. When incorporating all four features, the prediction accuracy was about 90%, indicating its desirable performance in distinguishing lodging from non-lodging plots. The model prediction accuracy of using all four features is not significantly different from that of using only two factors (i.e., texture and NDVI). Since data collection and processing workload increased with more features, researchers in the future could specifically focus on extracting and using texture, and NDVI features to train an SVM model for wheat lodging detection, instead of using four features (i.e., color, texture, NDVI, and height).