Importance of spatial and spectral data reduction in the detection of internal defects in food products

Experimental data manipulations to assess performance of hyperspectral classification models of crop seeds and other objects

BACKGROUND

Optical sensing solutions are being developed and adopted to classify a wide range of biological objects, including crop seeds. Performance assessment of optical classification models remains both a priority and a challenge.

METHODS

As training data, we acquired hyperspectral imaging data from 3646 individual tomato seeds (germination yes/no) from two tomato varieties. We performed three experimental data manipulations: (1) Object assignment error: effect of individual object in the training data being assigned to the wrong class. (2) Spectral repeatability: effect of introducing known ranges (0-10%) of stochastic noise to individual reflectance values. (3) Size of training data set: effect of reducing numbers of observations in training data. Effects of each of these experimental data manipulations were characterized and quantified based on classifications with two functions [linear discriminant analysis (LDA) and support vector machine (SVM)].

RESULTS

For both classification functions, accuracy decreased linearly in response to introduction of object assignment error and to experimental reduction of spectral repeatability. We also demonstrated that experimental reduction of training data by 20% had negligible effect on classification accuracy. LDA and SVM classification algorithms were applied to independent validation seed samples. LDA-based classifications predicted seed germination with RMSE = 10.56 (variety 1) and 26.15 (variety 2), and SVM-based classifications predicted seed germination with RMSE = 10.44 (variety 1) and 12.58 (variety 2).

CONCLUSION

We believe this study represents the first, in which optical seed classification included both a thorough performance evaluation of two separate classification functions based on experimental data manipulations, and application of classification models to validation seed samples not included in training data. Proposed experimental data manipulations are discussed in broader contexts and general relevance, and they are suggested as methods for in-depth performance assessments of optical classification models.

Root-associated entomopathogenic fungi manipulate host plants to attract herbivorous insects

Root-associated entomopathogenic fungi (R-AEF) indirectly influence herbivorous insect performance. However, host plant-R-AEF interactions and R-AEF as biological control agents have been studied independently and without much attention to the potential synergy between these functional traits. In this study, we evaluated behavioral responses of cabbage root flies [Delia radicum L. (Diptera: Anthomyiidae)] to a host plant (white cabbage cabbage Brassica oleracea var. capitata f. alba cv. Castello L.) with and without the R-AEF Metarhizium brunneum (Petch). We performed experiments on leaf reflectance, phytohormonal composition and host plant location behavior (behavioral processes that contribute to locating and selecting an adequate host plant in the environment). Compared to control host plants, R-AEF inoculation caused, on one hand, a decrease in reflectance of host plant leaves in the near-infrared portion of the radiometric spectrum and, on the other, an increase in the production of jasmonic, (+)-7-iso-jasmonoyl-l-isoleucine and salicylic acid in certain parts of the host plant. Under both greenhouse and field settings, landing and oviposition by cabbage root fly females were positively affected by R-AEF inoculation of host plants. The fungal-induced change in leaf reflectance may have altered visual cues used by the cabbage root flies in their host plant selection. This is the first study providing evidence for the hypothesis that R-AEF manipulate the suitability of their host plant to attract herbivorous insects.

Snapshot spectral polarimetric light field imaging using a single detector

In this Letter, we investigate a snapshot spectral-polarimetric-volumetric imaging (SSPVI) system using a single detector. Through compressed acquisition and reconstruction, SSPVI can achieve spectral imaging (x,y,λ), polarization imaging (x,y,ψ,χ), and light field imaging (x,y,θ,φ) simultaneously. The newly discovered performance is showcased by attaining the spectral-polarimetric-volumetric video and different laboratory accuracy experiments. These never-seen-before capacities of the camera open new prospects for many applications, such as biological analysis, object recognition, and remote sensing.

Non-destructive analysis of germination percentage, germination energy and simple vigour index on wheat seeds during storage by Vis/NIR and SWIR hyperspectral imaging

Two hyperspectral imaging (HSI) systems, visible/near infrared (Vis/NIR, 304-1082 nm) and short wave infrared (SWIR, 930-2548 nm), were used for the first time to comprehensively predict the changes in quality of wheat seeds based on three vigour parameters: germination percentage (GP, reflecting the number of germinated seedling), germination energy (GE, reflecting the speed and uniformity of seedling emergence), and simple vigour index (SVI, reflecting germination percentage and seedling weight). Each sample contained a small number of wheat seeds, which were obtained by high temperature and humidity-accelerated aging (0, 2, and 3 days) to simulate storage. The spectra of these samples were collected using HSI systems. After collection, each seed sample underwent a standard germination test to determine their GP, GE, and SVI. Then, several pretreatment methods and the partial least-squares regression algorithm (PLS-R) were used to establish quantitative models. The models for the Vis/NIR region obtained excellent performance, and most effective wavelengths (EWs) were selected in the Vis/NIR region by the successive projections algorithm (SPA) and regression coefficients (RC). Subsequently, PLS-R-RC models using selected wavebands (sixteen wavebands for GP, 14 wavebands for GE, and 16 wavebands for SVI) exhibited similar performance to the PLS-R models based on the full wavebands. The best R² results obtained in the simplified models' prediction sets were 0.921, 0.907, and 0.886, with RMSE values of 4.113%, 5.137%, and 0.024, for GP, GE, and SVI, respectively. Distribution maps of GP, GE, and SVI were produced by applying these simplified PLS models. By interpreting the EWs and building prediction models, soluble protein and sugar content were demonstrated to have a relationship with spectral information. In summary, the present results lay a foundation towards the development of a significantly simpler, more comprehensive, and non-destructive hyperspectral-based sorting system for determining the vigour of wheat seeds.

Convolutional neural network target detection in hyperspectral imaging for maritime surveillance

This work addresses a hyperspectral imaging system for maritime surveillance using unmanned aerial vehicles. The objective was to detect the presence of vessels using purely spatial and spectral hyperspectral information. To accomplish this objective, we implemented a novel 3-D convolutional neural network approach and compared against two implementations of other state-of-the-art methods: spectral angle mapper and hyperspectral derivative anomaly detection. The hyperspectral imaging system was developed during the SUNNY project, and the methods were tested using data collected during the project final demonstration, in São Jacinto Air Force Base, Aveiro (Portugal). The obtained results show that a 3-D CNN is able to improve the recall value, depending on the class, by an interval between 27% minimum, to a maximum of over 40%, when compared to spectral angle mapper and hyperspectral derivative anomaly detection approaches. Proving that 3-D CNN deep learning techniques that combine spectral and spatial information can be used to improve the detection of targets classification accuracy in hyperspectral imaging unmanned aerial vehicles maritime surveillance applications.

Improved spatial ecological sampling using open data and standardization: an example from malaria mosquito surveillance

Vector-borne disease control relies on efficient vector surveillance, mostly carried out using traps whose number and locations are often determined by expert opinion rather than a rigorous quantitative sampling design. In this work we propose a framework for ecological sampling design which in its preliminary stages can take into account environmental conditions obtained from open data (i.e. remote sensing and meteorological stations) not necessarily designed for ecological analysis. These environmental data are used to delimit the area into ecologically homogeneous strata. By employing Bayesian statistics within a model-based sampling design, the traps are deployed among the strata using a mixture of random and grid locations which allows balancing predictions and model-fitting accuracies. Sample sizes and the effect of ecological strata on sample sizes are estimated from previous mosquito sampling campaigns open data. Notably, we found that a configuration of 30 locations with four households each (120 samples) will have a similar accuracy in the predictions of mosquito abundance as 200 random samples. In addition, we show that random sampling independently from ecological strata, produces biased estimates of the mosquito abundance. Finally, we propose standardizing reporting of sampling designs to allow transparency and repetition/re-use in subsequent sampling campaigns.

Proximal Remote Sensing to Non-destructively Detect and Diagnose Physiological Responses by Host Insect Larvae to Parasitism

As part of identifying and characterizing physiological responses and adaptations by insects, it is paramount to develop non-destructive techniques to monitor individual insects over time. Such techniques can be used to optimize the timing of when in-depth (i.e., destructive sampling of insect tissue) physiological or molecular analyses should be deployed. In this article, we present evidence that hyperspectral proximal remote sensing can be used effectively in studies of host responses to parasitism. We present time series body reflectance data acquired from individual soybean loopers (Chrysodeixis includens) without parasitism (control) or parasitized by one of two species of parasitic wasps with markedly different life histories: Microplitis demolitor, a solitary larval koinobiont endoparasitoid and Copidosoma floridanum, a polyembryonic (gregarious) egg-larval koinobiont endoparasitoid. Despite considerable temporal variation in reflectance data 1-9 days post-parasitism, the two parasitoids caused uniquely different host body reflectance responses. Based on reflectance data acquired 3-5 days post-parasitism, all three treatments (control larvae, and those parasitized by either M. demolitor or C. floridanum) could be classified with >85 accuracy. We suggest that hyperspectral proximal imaging technologies represent an important frontier in insect physiology, as they are non-invasive and can be used to account for important time scale factors, such as: minutes of exposure or acclimation to abiotic factors, circadian rhythms, and seasonal effects. Although this study is based on data from a host-parasitoid system, results may be of broad relevance to insect physiologists. Described approaches provide a non-invasive and rapid method that can provide insights into when to destructively sample tissue for more detailed mechanistic studies of physiological responses to stressors and environmental conditions.

The potential and prospects of proximal remote sensing of arthropod pests

Bench-top or proximal remote sensing applications are widely used as part of quality control and machine vision systems in commercial operations. In addition, these technologies are becoming increasingly important in insect systematics and studies of insect physiology and pest management. This paper provides a review and discussion of how proximal remote sensing may contribute valuable quantitative information regarding identification of species, assessment of insect responses to insecticides, insect host responses to parasitoids and performance of biological control agents. The future role of proximal remote sensing is discussed as an exciting path for novel paths of multidisciplinary research among entomologists and scientists from a wide range of other disciplines, including image processing engineers, medical engineers, research pharmacists and computer scientists. © 2015 Society of Chemical Industry.

Remote Sensing and Reflectance Profiling in Entomology

Remote sensing describes the characterization of the status of objects and/or the classification of their identity based on a combination of spectral features extracted from reflectance or transmission profiles of radiometric energy. Remote sensing can be benchtop based, and therefore acquired at a high spatial resolution, or airborne at lower spatial resolution to cover large areas. Despite important challenges, airborne remote sensing technologies will undoubtedly be of major importance in optimized management of agricultural systems in the twenty-first century. Benchtop remote sensing applications are becoming important in insect systematics and in phenomics studies of insect behavior and physiology. This review highlights how remote sensing influences entomological research by enabling scientists to nondestructively monitor how individual insects respond to treatments and ambient conditions. Furthermore, novel remote sensing technologies are creating intriguing interdisciplinary bridges between entomology and disciplines such as informatics and electrical engineering.

Detection of temporal changes in insect body reflectance in response to killing agents

Computer vision and reflectance-based analyses are becoming increasingly important methods to quantify and characterize phenotypic responses by whole organisms to environmental factors. Here, we present the first study of how a non-destructive and completely non-invasive method, body reflectance profiling, can be used to detect and time stress responses in adult beetles. Based on high-resolution hyperspectral imaging, we acquired time series of average reflectance profiles (70 spectral bands from 434-876 nm) from adults in two beetle species, maize weevils (Sitophilus zeamais) and larger black flour beetles (Cynaus angustus). For each species, we acquired reflectance data from untreated controls and from individuals exposed continuously to killing agents (an insecticidal plant extract applied to maize kernels or entomopathogenic nematodes applied to soil applied at levels leading to ≈100% mortality). In maize weevils (exposed to hexanic plant extract), there was no significant effect of the on reflectance profiles acquired from adult beetles after 0 and 12 hours of exposure, but a significant treatment response in spectral bands from 434 to 550 nm was detected after 36 to 144 hours of exposure. In larger black flour beetles, there was no significant effect of exposure to entomopathogenic nematodes after 0 to 26 hours of exposure, but a significant response in spectral bands from 434-480 nm was detected after 45 and 69 hours of exposure. Spectral bands were used to develop reflectance-based classification models for each species, and independent validation of classification algorithms showed sensitivity (ability to positively detect terminal stress in beetles) and specificity (ability to positively detect healthy beetles) of about 90%. Significant changes in body reflectance occurred at exposure times, which coincided with published exposure times and known physiological responses to each killing agent. The results from this study underscore the potential of hyperspectral imaging as an approach to non-destructively and non-invasively quantify stress detection in insects and other animals.