•The YOLOv7 model was improved to detect small and indistinguishable targets.•The Mean Average Precision and F1-Score metrics were improved by adding two CBAM modules.•The model size and ...computational cost were decreased by replacing standard convolution layers in all the ELAN, ELAN_W, and SPPCSPC modules with ghost modules.•The detected objects were tracked and counted using the ByteTrack multi-object tracking and line-cross algorithm, respectively.
The occurrence of heated and immature canola kernels caused by excessive drying and frost damage is undesired by grain buyers due to lower oil yield and diminished market value. The current grading process is visually examining each kernel's endosperm color and counting the damaged seeds. As this process is time-consuming, laborious, and prone to errors, this study proposes an automated grading technique based on object detection, multi-object tracking, and counting. The detection task was achieved via an improved YOLOv7 network (YOLOv7_ours) that was modified to increase its performance in accurately identifying small objects by adding two convolutional block attention modules in the neck region and decrease its computational complexity (cost) and size by substituting convolutional layers with ghost layers in all the Efficient Layer Aggregation Networks modules, and in the Spatial Pyramid Pooling Cross Stage Partial module present in YOLOv7. The weights of the trained network were fed to the ByteTrack multiple object tracker to track the detections frame-by-frame in a video feed. The unique identities generated by the tracker for each detected object of interest were then used to count the number of defects using a line cross algorithm. The mean average precision (mAP@0.5) obtained after training the YOLOv7_ours model was 1.02% better and its cost and size were 32.1% and 37.1% lower than the baseline YOLOv7 model. In a test video, the overall model achieved a multi-object tracking accuracy and counting accuracy of 84.8% and 93.9%, respectively. This three-stage model can be readily deployed in an edge device for accurate and real-time grading of canola kernels by grain buyers.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
•Smart-portable VNIR hyperspectral imaging was used for plant-based meat color prediction.•CIE-L*, a*, b* color space of burgers was unstable, even over a short storage period.•VNIR spectral features ...revealed the non-uniform surface distribution of PB color.•Feature wavelength selection with B-iPLS and GAs improved the performance of PLSR prediction models.•Pixel-by-pixel prediction maps showed distinct color degradation patterns.
The excessive color variation in plant-based meat burgers (PB) during storage, arising from the interplay of multiple ingredients, suggests the acute need for real-time color monitoring techniques. This study reports the use of a smart-portable hyperspectral imaging device in the visible-near-infrared coupled with multivariate data and image analysis for the prediction of color in PB samples of different formulations during storage. Principal Components Analysis (PCA) and Partial Least Squares Regression (PLSR) were applied to explore the spectral features and predict the CIE - L*, a*, b*. PCA on spectral data showed successful spatial separation of PB samples based on color stability. The PLSR models in the full spectral range possessed good prediction accuracies for L* (R2P = 0.88, RMSEP = 0.94), a* (R2P = 0.98, RMSEP = 0.15), b* (R2P = 0.92, RMSEP = 0.31). Prediction capabilities of simplified PLSR models were improved for a* (R2P = 0.99, RMSEP = 0.09) with B-iPLS, b* (R2P = 0.95, RMSEP = 0.23) with B-iPLS. Pixel-by-pixel prediction maps unveiled the pattern of color deterioration concerning individual base ingredients in the PB samples. The results unravel the feasibility of developing a real-time multi-spectral technique for quantification and visualization of color evolution in plant-based meat analogs.
Display omitted
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
•Designed a comprehensive computer vision system for canola seed damage identification.•Evaluated manual and semi-automatic sample preparation methods in terms of dataset quality and time ...efficiency.•Substituted CSPDarkNet53 with lightweight CNNs (ShuffleNetV2 and MobileNetV3) in YOLOv5s for size and computational complexity reduction.•Exported ShuffleNetV2_yolov5s to ONNX format and deployed it on an NVIDIA Jetson Nano, integrated with the semi-automatic machine.
The current method employed in detecting damages present in canola kernels is inefficient and time-consuming. The sample preparation process is laborious and the judgment between sound and damaged seeds using visual methods is prone to errors as they are small and indistinguishable. To date, there is no hardware and supporting software solution that can expedite this time-consuming yet crucial task. This study demonstrates an end-to-end damage detection framework that has two components: a semi-automatic machine to crush the canola seeds to prepare the samples rapidly for detecting the damages and a supporting object detection algorithm based on a compressed YOLOv5s model. The lightweight detection model was deployed in an Edge-AI device and integrated with the crusher to detect the damages in real-time. Two Convolutional Neural Network (CNN) architectures, viz. ShuffleNetV2 and MobileNetV3 were compared in terms of model parameters, computational cost, and model size to replace the standard CSPDarknet53 CNN backbone present in YOLOv5s. The best-performing model was deployed into an NVIDIA Jetson Nano embedded device for real-time inferencing. Results indicate that in terms of model size, the ShuffleNetV2_YOLOv5s model was 56.5 and 80.4 % smaller than the MobileNetV3_YOLOv5s and the baseline YOLOv5s models, respectively. While, in terms of computational cost, the ShuffleNetV2_YOLOv5s model was 64.1 % and 99.5 % less expensive than the MobileNetV3_YOLOv5s and the baseline YOLOv5s models, respectively. This study demonstrated an end-to-end framework for detecting damages in canola supported by a real-time and lightweight damage detection model.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Non-destructive quality evaluation techniques are vital in 3D visualization and quantification of microstructural changes in micronized pulse kernels. The effects of micronization (infrared heating) ...on pulse microstructure are highly dependent on its treatment efficiency. This study employed X-ray micro-computed tomography (micro-CT) to characterize, visualize, and quantify the microstructural changes non-destructively. Four major pulses, chickpeas, fava beans, lentils, and yellow peas, were subjected to micronization at 180 °C of surface temperature for four different infrared (IR) exposure times (60, 80, 100, and 120 s). All samples were pre-conditioned at 20 % moisture content before the micronization process. Morphometric parameters (pore size, diameter, and pore distribution) were qualitatively measured using the reconstructed 2D and 3D images acquired by the X-ray micro-CT. Physico-chemical parameters, including loss of moisture (%), total porosity (%), volumetric expansion/shrinkage (%), and density (%) were quantified using Micro-CTAn software. The results showed that the porosity is significantly increased (p < 0.05) in all pulse types ranging from 0.60 ± 0.42 % to 12.1 ± 2.68 %, at prolonged IR exposure time. Significant volume expansion ranging from 3.82 %–33.56 % was shown by chickpeas and yellow peas, while fava beans and lentils showed volume shrinkage ranging from 4.26 %–24.31 % after micronization. Furthermore, density was decreased ranging from 0.79 ± 0.25% to 33.49 ± 0.42 % with increased IR time for all pulses except fava beans. The authors found that the microstructural changes in all tested pulses were optimum at 100 s–120 s of exposure time. Results emphasize the higher potential in modifying the microstructure of pulses using micronization processing. Therefore, future studies are suggested on developing predictive models to explore the relationships between microstructural changes and techno-functional properties.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
•The optical properties of Braeburn apples before and after shelf-life were measured with a setup for spatially resolved diffuse reflectance spectroscopy.•The estimated absorption spectra showed good ...correlation with the SSC.•The estimated absorption spectra gave better prediction of the soluble solids content and firmness than the scattering spectra.•The diffuse reflectance spectra gave comparable prediction accuracies to the μa spectra.
Spectroscopic measurements in the visible and near infrared wavelength range have achieved success in non-destructive assessment of apple quality attributes contributed by chemical components inside the fruit such as sweetness. Nevertheless, the evaluation of quality attributes related to texture of the fruit (e.g. firmness) still remains a challenge. One of the proposed solutions is to acquire and utilize separately scattering and absorption information from spectroscopic readings for quality prediction. Since scattering is related to fruit microstructure and absorption is caused by chemical composition, construction of multivariate calibration models from these optical properties to predict corresponding quality parameters of interest could be a solution for this, and also have potential for non-destructive monitoring of fruit quality from harvest to consumption. In this research, a setup for contact spatially resolved diffuse reflectance measurements in the 500–1000nm range based on a fibre-optics probe was elaborated for the measurement of optical properties (absorption coefficient μa and reduced scattering coefficient μ′s) of ‘Braeburn’ apples. After calibration and validation of the setup on a set of liquid optical phantoms covering the relevant range of optical properties, thirty ‘Braeburn’ apples were measured before and after shelf-life storage (2 weeks at 18°C in normal atmosphere) with the setup and were analyzed for the main quality attributes (firmness and soluble solids content (SSC)). The estimated μa spectra of the apples indicated chlorophyll degradation during shelf-life storage. PLS models were investigated for apple quality prediction by using estimated optical properties spectra or diffuse reflectance spectra. These spectra covered information on chlorophyll and some carbohydrate and water absorption. The μa spectra also proved better than μ′s spectra for predicting SSC and firmness (R2-SSC=0.81; RMSEP(SSC)=0.69%; and R2-firmness=0.71; RMSEP(firmness)=9.68N). The combined μa and μ′s spectra did not improve the prediction accuracies as compared to the μa spectra alone. The diffuse reflectance spectra of the detection fibres did not provide a significantly better prediction performance for SSC, but gave slightly better firmness prediction (R2-firmness=0.73–0.83; RMSEP(firmness)=8.91–13.70N) than the μa spectra.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK
With significant interest in incorporating beans, lentils, and pulses as nutrient-rich healthy food sources into our diets, a reliable technique for their rapid and accurate quality evaluation is ...needed. The method of single kernel characterization to determine the physical properties (i.e., diameter, weight, moisture content, and hardness) of mung beans was assessed in this study. Two mung bean varieties were characterized using the single kernel characterization technique and the results were compared to traditional methods. It was observed that predicted bean weights were accurate to known laboratory measurements (
R =
0.98,
n
= 200). Individual bean characterization was moderately (
R =
0.58–0.7,
n =
100) correlated in regard to the true diameter of mung beans. An evaluation on moisture content was performed after tempering the two bean varieties to four moisture levels and a good correlation was obtained with the oven drying method (
R
= 0.92,
n
= 24). Hardness values obtained by single kernel characterization were moderately correlated to maximum forces measured using an Instron universal testing system. However, a common relationship was observed between mung bean hardness and moisture content when using both methods. Compared to visual inspection, automated characterization of single beans is a superior technique to measure the geometrical and mechanical properties of mung beans in an industrial setup where high throughput is paramount.
Full text
Available for:
EMUNI, FIS, FZAB, GEOZS, GIS, IJS, IMTLJ, KILJ, KISLJ, MFDPS, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, SBMB, SBNM, UKNU, UL, UM, UPUK, VKSCE, ZAGLJ
The storage potential of apples highly depends on the maturity at harvest. Optical methods have been proposed to measure maturity in a fast, reliable and non-destructive way. However, the signal is ...often composed of photons with different penetration depths into the material. An attempt to separate these photons might result in more precise correlations with quality attributes, as these could relate to a specific layer/depth into the sample. Therefore, a Vis/NIR spatially resolved laser reflectance setup was used, combining a supercontinuum laser and a monochromator to illuminate samples with a monochrome focused beam in the 550–1000 nm wavelength range. A panchromatic camera was used to obtain diffuse reflectance profiles at each wavelength. In the period starting 50 days before until 11 days after commercial harvest, 320 Braeburn apples were measured. Partial least squares regression models were developed to relate apple maturity/quality to the diffuse reflectance spectra at different distances from the illumination point. The effect of detector size (spatial bandwidth) was also evaluated. A bandwidth of 0.82 mm in combination with a parameter specific illumination-detection distance, gave the best results. Using an internal test set, an R ² of prediction of 0.98 and 0.93, and a ratio of prediction to performance (RPD) of 5.84 and 3.42, predicting, respectively, the Streif index and starch conversion values was obtained. The predictions of soluble solids content (SSC) (R ² of 0.81; RPD of 2.04) and firmness (R ² of 0.65; RPD of 1.66) were less accurate. Also, worse predictions were obtained using an external test set.
Full text
Available for:
EMUNI, FIS, FZAB, GEOZS, GIS, IJS, IMTLJ, KILJ, KISLJ, MFDPS, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, SBMB, SBNM, UKNU, UL, UM, UPUK, VKSCE, ZAGLJ
In many research areas and application domains, the bulk optical properties of biological materials are of great interest. Unfortunately, these properties cannot be obtained easily for complex turbid ...media. In this study, a metamodeling approach has been proposed and applied for the fast and accurate estimation of the bulk optical properties from contactless and non-destructive hyperspectral scatter imaging (HSI) measurements. A set of liquid optical phantoms, based on intralipid, methylene blue and water, were prepared and the Vis/NIR bulk optical properties were characterized with a double integrating sphere and unscattered transmittance setup. Accordingly, the phantoms were measured with the HSI technique and metamodels were constructed, relating the Vis/NIR reflectance images to the reference bulk optical properties of the samples. The independent inverse validation showed good prediction performance for the absorption coefficient and the reduced scattering coefficient, with R(2)(p) values of 0.980 and 0.998, and RMSE(P) values of 0.032 cm(-1) and 0.197 cm(-1) respectively. The results clearly support the potential of this approach for fast and accurate estimation of the bulk optical properties of turbid media from contactless HSI measurements.
A metamodeling approach is introduced and applied to efficiently estimate the bulk optical properties of turbid media from spatially resolved spectroscopy (SRS) measurements. The model has been ...trained on a set of liquid phantoms covering a wide range of optical properties representative for food and agricultural products and was successfully validated in forward and inverse mode on phantoms not used for training the model. With relative prediction errors of 10% for the estimated bulk optical properties the potential of this metamodeling approach for the estimation of the optical properties of turbid media from spatially resolved spectroscopy measurements has been demonstrated.