Maize kernels are easily infected with different kinds of moulds in nature which has terrible negative effects on kernels. In this research, a series of multi-channel residual modules (MCRMs) were introduced to convolutional neural network (CNN) to identify the mould varieties combined with Raman hyperspectral imaging technique. Specifically, Raman hyperspectral images of maize kernels infected with different moulds were acquired and their characteristics were summarized and prepared for further analysis. Traditional machine learning (TML) models and deep learning (DL) models were both established and compared. For TML, three kinds of modeling methods combined with four kinds of variables selection methods were applied to establish the mould varieties identification model. For DL, five different residual units were explored to obtain the optimal MCRM-CNN architecture. The results showed that MCRM-CNN was more capable of mining hidden information compared with TML methods. Finally, MCRM-CNN-SVM provided the optimal model to identify the moulds varieties with accuracy in the testing set of 100% by taking the variables extracted by MCRM-CNN as the inputs of SVM. The residual unit was composed of three stacking residual blocks with different kernels of 3, 5, 7 without the shortcut connection (MNR) and the recall rates of A. niger, A. flavus, A. glaucus, A. parasiticus and control set were all 100%, respectively. Consequently, the proposed method expanded the nondestructive online detection for identifying mould varieties infecting maize kernels by combining RHSI and DL.

Maize (Zea mays L.) holds a pivotal position in various domains, making maize seeds' germination rate and quality crucial for agricultural production. Insect -infested seeds can impair their germination rate, resulting in substantial economic losses. Nevertheless, research on rapid detection methods for insect infestation in maize seeds remains insufficient. In this study, we employed hyperspectral data within the range of 930-1866 nm alongside intelligent algorithms to address the classification of healthy and insect -infested maize seeds. The 1DCNN-BiLSTM and SVM models were applied to diverse combinations of spectra and texture features. The results indicated that the 1D-CNN-BiLSTM model exhibited clear advantages over SVM in handling multi -source data. Additionally, ANOVA identified the 1160 nm / 1310 nm band ratio as optimal. The 1D-CNN-BiLSTM model, utilizing texture features from this band ratio image, exhibited exceptional classification performance with just two band images. The GLCM + 1D-CNN-BiLSTM model garnered the best results, with the F1 -score and accuracy of 0.96, respectively. This approach streamlined the band number, reducing model complexity and rendering it promising for practical applications. Consequently, the results of this study provide new insights and methods for classifying insect infestation maize seeds.

The pulse-width modulation (PWM) variable spray system is the most widely used variable spray system in the world at present, which has the characteristics of a fast response, large flow adjustment range, and good atomization. Recently, the pressure fluctuation and droplet deposition uniformity of the PWM variable spray system caused by the intermittent spray mode of the nozzle have attracted more and more attention. In this study, a method for eliminating the inhomogeneity of ground deposition in low-frequency PWM variable sprays based on a staggered-phase drive mode was proposed, and a PWM variable spray system was built. The experimental results indicated that the pressure fluctuation amplitude upstream of the nozzle of the PWM variable spray system with the staggered-phase drive was reduced by 40.91%, and the dispersion rate of the pressure fluctuation was reduced by 62.78% (the initial pressure was 0.3 MPa, solenoid valve frequency was 5 Hz, and duty cycle was 50%). The PWM control parameters had a significant effect on the upstream pressure fluctuation (initial pressure > duty cycle > frequency). The droplet spectrum relative span of the staggered phased PWM variable spray system decreased by 24.83%, the coefficient of variation of the droplet particle size decreased by 4.40%, the particle size was more uniform, and the atomization effect was improved. The average deposition of droplets in the forward direction driven by the staggered phase was 4.87% greater than that in the same phase, and the variation rate decreased by 20.87%. The average deposition amount increased, and the deposition became more uniform. Staggered-phase spray control could effectively reduce the inhomogeneity of deposition in low-frequency PWM intermittent spraying. This research provides strong technical support for a precision variable spraying effect and droplet drift prevention.

Moisture content (MC) is one of the important indexes to evaluate maize seed quality. Its accurate prediction is very challenging. In this study, the long-wave near-infrared hyperspectral imaging (LW-NIR-HSI) system was used, and the embryo side (S1) and endosperm side (S2) spectra of each maize seed were extracted, as well as the average spectrum (S3) of both being calculated. The partial least square regression (PLSR) and least-squares support vector machine (LS-SVM) models were established. The uninformative variable elimination (UVE) and successive projections algorithm (SPA) were employed to reduce the complexity of the models. The results indicated that the S3-UVE-SPA-PLSR and S3-UVE-SPA-LS-SVM models achieved the best prediction accuracy with an RMSEP of 1.22% and 1.20%, respectively. Furthermore, the combination (S1+S2) of S1 and S2 was also used to establish the prediction models to obtain a general model. The results indicated that the S1+S2-UVE-SPA-LS-SVM model was more valuable with Rpre of 0.91 and RMSEP of 1.32% for MC prediction. This model can decrease the influence of different input spectra (i.e., S1 or S2) on prediction performance. The overall study indicated that LW-HSI technology combined with the general model could realize the non-destructive and stable prediction of MC in maize seeds.

BACKGROUND: Bruises caused by mechanical collision during the harvesting and storage and transportation period are difficult to detect using traditional machine vision technologies because there is no obvious difference in appearance between bruised and sound tissues. As a result of its fast and non-destructive characteristics, hyperspectral imaging technology is a potential tool for non-destructive detection of fruit surface defects. RESULTS: In the present study, visible near infrared hyperspectral reflectance images of healthy apples and bruised apples at 6, 12 and 24 h were obtained. To reduce hyperspectral data dimension, optimal wavelength selection algorithms including principal component analysis (PCA) and band ratio methods were utilized to select the effective wavelengths and enhance the contrast between bruised and sound tissues. Then pseudo-color image transformation technology combining with improved watershed segmentation algorithm (IWSA) were employed to recognize the bruise spots. The result obtained showed that band ratio images obtained better detection performance than that of PCA. The G component derived from pseudo-color image of o.821-.752THORN=o. 821 +.752THORN followed by IWSA obtained the best segmentation performance for bruise spots. Finally, a multispectral imaging system for the detection of bruised apple was developed to verify the effectiveness of the proposed two-band ratio algorithm, obtaining recognition rates of 93.3%, 92.2% and 92.5% for healthy, bruised and overall apples, respectively. CONCLUSION: The bruise detection algorithm proposed in the present study has potential to detect bruised apple in online practical applications and hyperspectral reflectance imaging offers a useful reference for the detection of surficial defects of fruit. (c) 2023 Society of Chemical Industry.

Seed vigor is one of the essential contents of agricultural research. The decline of seed vigor is described as an inevitable process. Recent studies have shown that the oxidative damage caused by reactive oxygen species (ROS) is the main reason for the destruction of various chemicals in seeds and eventually evolves into seed death. The traditional vigor tests, such as the seed germination test and TTC staining, are commonly used to assess seed vigor. However, these methods often need a large number of experimental samples, which will bring a waste of seed resources. At present, many new methods that are fast and nondestructive to seeds, such as vibrational spectroscopic techniques, have been used to test seed vigor and have achieved convincing results. This paper is aimed at analyzing the microchanges of seed-vigor decline, summarizing the performance of current seed-vigor test methods, and hoping to provide a new idea for the nondestructive testing of a single seed vigor by combining the physiological alterations of seeds with chemometrics algorithms.

At present, the apple grading system usually conveys apples by a belt or rollers. This usually leads to low hardness or expensive fruits being bruised, resulting in economic losses. In order to realize real-time detection and classification of high-quality apples, separate fruit trays were designed to convey apples and used to prevent apples from being bruised during image acquisition. A semantic segmentation method based on the BiSeNet V2 deep learning network was proposed to segment the defective parts of defective apples. BiSeNet V2 for apple defect detection obtained a slightly better result in MPA with a value of 99.66%, which was 0.14 and 0.19 percentage points higher than DAnet and Unet, respectively. A model pruning method was used to optimize the structure of the YOLO V4 network. The detection accuracy of defect regions in apple images was further improved by the pruned YOLO V4 network. Then, a surface mapping method between the defect area in apple images and the actual defect area was proposed to accurately calculate the defect area. Finally, apples on separate fruit trays were sorted according to the number and area of defects in the apple images. The experimental results showed that the average accuracy of apple classification was 92.42%, and the F1 score was 94.31. In commercial separate fruit tray grading and sorting machines, it has great application potential.

针对算力有限的移动和嵌入式平台,提出了一种基于深度学习的轻量化火焰烟雾检测算法.利用数据增强来解决数据量较少的问题,使用one-stage目标检测方法中的YOLOv4作为火焰烟雾检测的模型框架,采用轻量化神经网络MobileNetV3替换YOLOv4的原主干特征提取网络,减少了模型参数量;再采用深度可分离卷积替换掉YOLOv4中的标准卷积块,进一步在加强特征提取网络和预测层减少了参数量;最后对空间金字塔池化部分进行改进,减少背景干扰带来的影响,减少最大池化导致的部分有用特征信息丢失.在该数据集上通过与原网络模型和其他主流目标检测方法进行对比分析,结果表明提出的轻量化网络不但保留了原模型精度,还大大减小了网络的训练参数量,提高了运行速度,更有利于模型搭载在摄像头等嵌入式设备上,实现火焰和烟雾的实时检测.

Vigor is an important indicator for seed quality evaluation. In this study, the mechanical process of loss of maize seed vigor was investigated based on the micro-fluorescence technique. Seeds with ten aging gradients were obtained with the temperature of 45 °C and humidity of 100%. The seeds without aging treatment were as controls. Three common vigor assays were used to evaluate seed vigor including germination test, TTC staining, and conductivity test. Significant changes in starch content were found in the seeds with aging treatments from macroscopic and microscopic perspectives. The findings of the conductivity vigor assay were validated by fluorescent imaging of proteins to further show the process of membrane structural changes during seed vigor declining. Scanning electron microscopy (SEM) experiments were used for the structural observation of starch granules, and the results showed the transformation of starch granules from small spherical to polyhedral shapes. In addition, the micro-fluorescence spectra of the sections showed several characteristic peaks related to vigor. This study was focused on the mechanism of seed aging and provided a new perspective and theoretical basis for seed vigor with artificial aging treatment.

为解决番茄缺陷检测过程中的精确性和实时性问题,该研究提出一种基于模型剪枝的番茄表面缺陷实时检测方法.采用模型剪枝的方法在YOLOv4网络模型基础上进行模型优化,首先将3个连续检测工位采集的RGB图像拼接生成YOLOv4网络的输入图像,然后采用通道剪枝和层剪枝的方法压缩YOLOv4网络模型,从而减少模型参数,提高检测速度,最后提出一种基于L1范数的非极大值抑制方法,用于在模型微调后去除冗余预测框,从而精准定位图像中的缺陷位置,并将模型部署到分级系统上进行实时检测试验.结果表明,该研究提出的YOLOv4P网络与原YOLOv4网络相比,网络模型尺寸和推理时间分别减少了232.40 MB和10.11 ms,平均精度均值(Mean Average Precision,mAP)从92.45％提高到94.56％,能满足实际生产中针对缺陷番茄进行精准、实时检测的要求,为番茄分级系统提供了高效的实时检测方法.

Slight crack of cottonseed is a critical factor influencing the germination rate of cotton due to foamed acid or water entering cottonseed through testa. However, it is very difficult to detect cottonseed with slight crack using common non-destructive detection methods, such as machine vision, optical spectroscopy, and thermal imaging, because slight crack has little effect on morphology, chemical substances or temperature. By contrast, the acoustic method shows a sensitivity to fine structure defects and demonstrates potential application in seed detection. This paper presents a novel method to detect slightly cracked cottonseed using air-coupled ultrasound with a light-weight vision transformer (ViT) and a sound-to-image encoding method. The echo signal of air-coupled ultrasound from cottonseed is obtained by non-contact and non-destructive methods. The intrinsic mode functions (IMFs) of ultrasound signal are obtained as the sound features using variational mode decomposition (VMD) approach. Then the sound features are converted into colorful images by a color encoding method. This method uses different colored lines to represent the changes of different values of IMFs according to the specified encoding period. A light-weight MobileViT method is utilized to identify the slightly cracked cottonseeds using encoding colorful images corresponding to cottonseeds. The experimental results show an average overall recognition accuracy of 90.7% for slightly cracked cottonseed from normal cottonseed, which indicates that the proposed method is reliable to applications in detection task of cottonseed with slight crack.

In order to realize on-line detection of defective apples on a two-lane fruit sorting machine, an inspection module was constructed using NIR cameras and a diffuse illumination chamber. A real-time apple defects inspection method was proposed based on YOLO V4 deep learning algorithm. The input images were generated by combining NIR images in three consecutive rubber roller stations. Channel pruning and layer pruning methods were used to simplify the YOLO V4 network and accelerate the detection speed. A non-maximum suppression (NMS) method based on Ll norm is proposed to remove redundant prediction box after fine-tuning the pruned network. The test results indicated that the model size and inference time of the pruning-based YOLO V4 network was decreased by 241.24 megabyte (MB) and 10.82 ms, respectively, and the mean average precision (mAP) was increased from 91.82% to 93.74%, compared with the YOLO V4 network before pruning. The pruning-based YOLO V4 network based on NIR images was not affected by the variation of skin color and suitable for detects identification of different cultivars including 'Fuji' apple covered in red-yellow striping and red blush, `Golden Delicious', and 'Granny Smith', with the average detection accuracy of 93.9% at the on-line test assessing five fruit per second. The overall results showed that the proposed pruning-based YOLO V4 network combined with the developed inspection module, had great potential to be implemented in commercial fruit packing line for fruit defects identification.

Maize is an important food crop in the world and it is used in many fields. The classification of maize seed maturity is of great value because it could increase the yield. In this study, near-infrared hyperspectral imaging (NIR-HSI) was employed to explore the maturity classification of maize seeds. In order to observe the influence of spectra of different positions in maize seed for modeling, the hyperspectral images of embryo and endosperm sides of maize seeds were collected in the spectral range of 1000-2300 nm. The average spectra of the embryo side (T1) and endosperm (T2) side were extracted from hyperspectral images, and then, the average spectra of both sides of maize seed (T3) were also calculated. T1, T2 and T3 spectra were used to build calibration models for maturity classification, respectively. And T1 and T2 spectra were imported into these developed classification models, and the classification accuracy of two types of spectra in the model was used to evaluate model applicability. These modeling methods including partial least square discriminant analysis (PLS-DA), decision tree (DT) and adaptive boosting (AdaBoost) methods. The principal component analysis (PCA) was applied to select feature wavelengths, common peaks and valleys in the loading curves of PC1 and PC2 were regarded as feature wavelengths. In order to reduce the influence of division of the calibration set, 50 randomized independent trials were carried out, and the average accuracy and stableness were used to evaluate the performance of models. Comparing among all models, PLS-DA model based on feature wavelengths selected by T2 spectra obtained the optimal results. When T1 and T2 were used as input to the optimal model, the classification accuracy was 98.7% and 100%, respectively. These results demonstrate the potential of the hyperspectral imaging technology for the rapid and accurate classification of maize seed maturity, and the feature wavelengths selected from the endosperm side combined with PLS-DA algorithm could establish a stable model.

Moisture content (MC) is one of the most important factors for assessment of seed quality. However, the accurate detection of MC in single seed is very difficult. In this study, single maize seed was used as research object. A long-wave near infrared (LWNIR) hyperspectral imaging system was developed for acquiring reflectance images of the embryo and endosperm side of maize seed in the spectral range of 930-2548 nm, and the mixed spectra were extracted from both side of maize seeds. Then, Full spectrum models were established and compared based on different types of spectra. It showed that models established based on spectra of the embryo side and mixed spectra obtained better performance than the endosperm side. Next, a combination of competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) was proposed to select the most effective wavelengths from full spectrum data. In order to explore the stableness of wavelength selection algorithm, these methods were used for 200 independent experiments based on embryo side and mixed spectra, respectively. Each selection result was used as input of partial least squares regression (PLSR) and least squares support vector machine (LS-SVM) to build calibration models for determining the MC of single maize seed. Results indicated that the CARS-SPA-LS-SVM model established with mixed spectra was optimal for MC prediction in all models by considering the accuracy, stableness and complexity of models. The prediction accuracy of CARS-SPA-LS-SVM model is R-pre = 0.9311 +/- 0.0094 and RMSEP = 1.2131 +/- 0.0702 in 200 independent assessment. The overall study revealed that the long-wave near infrared hyperspectral imaging can be used to non-invasively and fast measure the MC in single maize seed and a robust and accurate model could be established based on CARS-SPA-LS-SVM method coupled with mixed spectral. These results can provide a useful reference for assessment of other internal quality attributes (such as starch content) of single maize seed. (C) 2021 Elsevier B.V. All rights reserved.

As an essential factor in quality assessment of maize seeds, variety purity profoundly impacts final yield and farmers' economic benefits. In this study, a novel method based on Raman hyperspectral imaging system was applied to achieve variety classification of coated maize seeds. A total of 760 maize seeds including 4 different varieties were evaluated. Raman spectral data of 400-1800 cm-1 were extracted and preprocessed. Variable selection methods involved were modified competitive adaptive reweighted sampling (MCARS), successive projections algorithm (SPA), and their combination. In addition, MCARS was proposed for the first time in this paper as a stable search technology. The performance of support vector machine (SVM) models optimized by genetic algorithm (GA) was analyzed and compared with models based on random forest (RF) and back-propagation neural network (BPNN). Same models based on Vis-NIR spectral data were also established for comparison. Results showed that the MCARS-GA-SVM model based on Raman spectral data obtained the best performance with calibration accuracy of 99.29% and prediction accuracy of 100%, which were stable and easily replicated. In addition, the accuracy on the independent validation set was 96.88%, which proved that the model can be applied in practice. A more simplified MCARS-SPA-GA-SVM model, which contained only 3 variables, had more than 95% accuracy on each data set. This procedure can help to develop a real-time detection system to classify coated seed varieties with high accuracy, which is of great significance for assessing variety purity and increasing crop yield.

Decay is a serious problem in citrus storage and transportation. However, the automatic detection of decayed citrus remains a problem. In this study, the long wavelength near-infrared (LW-NIR) hyperspectra reflectance images (1000-1850 nm) of oranges were obtained, and an effective method to detect decayed citrus was proposed. Three effective wavelength selection algorithms and two classification algorithms were used to build decay detection models in pixel-level, as well as the two-band ratio images, pseudo-color image enhancement and improved watershed segmentation were used to build decay detection models in image-level. The imagelevel detection method proposed in this study obtained a total success rate of 92% for all fruit, indicating its potential to detect decayed oranges online. Moreover, the LW-NIR hyperspectral reflectance imaging is verified as a useful method to detect surface defects of fruits.

A deep-learning architecture based on Convolutional Neural Networks (CNN) and a cost-effective computer vision module were used to detect defective apples on a four-line fruit sorting machine at a speed of 5 fruits/s. A CNN based classification architecture was trained and tested, with the accuracy, recall, and specificity of 96.5%, 100.0%, and 92.9%, respectively, for the testing set. An inferior performance was obtained by a traditional image processing method based on candidate defective regions counting and a support vector machine (SVM) classifier, with the accuracy, recall, and specificity of 87.1%, 90.9%, and 83.3%, respectively. The CNN-based model was loaded into the custom software to validate its performance using independent 200 apples, obtaining an accuracy of 92% with a processing time below 72 ms for six images of an apple fruit. The overall results indicated that the proposed CNN-based classification model had great potential to be implemented in commercial packing line.

The moisture content (MC) of cucumber seeds was detected nondestructively using two hyperspectral imaging (HSI) systems with complementary spectral ranges. The mean spectrum of each cucumber seed was extracted from hyperspectral images in 400-1000 and 1050-2500 nm separately and it was found that the reflectance spectra decreased as the MC increased in 1050-2500 nm. Calibration models were established by partial least squares regression (PLS) to analyze the predictive ability of preprocessing and wavelength selection methods. The spectra in 400-1000 nm pretreated by Savitzky-Golay smoothing and standard normal variate (SG-SNV) and the 1050-2500 nm spectra pretreated by SG-normalization yielded better results. The optimal wavelengths were obtained by three effective wavelength selection methods, i.e., competitive adaptive reweighted sampling (CARS), iteratively retains informative variables (IRIV), and random frog (RF). Subsequently, the simplified models were built by the selected wavelengths separately. Compared to other developed models, the calibration model established with eight wavelengths chosen by RF from hyperspectral images at 1050-2500 nm achieved optimal performance. The correlation coefficient of prediction (R-pre) was 0.917 and the root mean square error of prediction (RMSEP) was 1.656%. Finally, the visualization of MC distribution was generated at the pixel level. The obtained results in this work indicated that applying HSI technology to measure MC in cucumber seeds was feasible, and the spectrum in 1050-2500 region was more promising than 400-1000 for MC detection. The visualization of MC distribution provided by HSI ensured comprehensive evaluation of MC in single seed level. The selected wavelengths were useful for building a multispectral imaging system to detect MC of cucumber seeds, which could get rid of the seeds with high MC and avoid seed deterioration during storage quickly.

Over the past decades, imaging and spectroscopy techniques have been developed rapidly with widespread applications in non-destructive agro-food quality determination. Seeds are one of the most fundamental elements of agriculture and forestry. Seed viability is of great significance in seed quality characteristics reflecting potential seed germination, and there is a great need for a quick and effective method to determine the germination condition and viability of seeds prior to cultivate, sale and plant. Some researches based on spectra and/or image processing and analysis have been explored in terms of the external and internal quality of a variety of seeds. Many attempts have been made in image segmentation and spectra correction methods to predict seed quality using various traditional and novel methods. This review focuses on the comparative introduction, development and applications of emerging techniques in the analysis of seed viability, in particular, near infrared spectroscopy, hyperspectral and multispectral imaging, Raman spectroscopy, infrared thermography, and soft X-ray imaging methods. The basic theories, principle components, relative chemometric processing, analytical methods and prediction accuracies are reported and compared. Additionally, on the foundation of the observed applications, the technical challenges and future outlook for these emerging techniques are also discussed.

The automated detection of defective apples with a machine vision system is difficult because of the non-uniform intensity distribution on the apple images and the visual similarity between the stem-ends/calyx and the defects. This paper presents a novel method to recognise defective apples by using a machine vision system that combines near-infrared(NIR) coded spot-array structured light and fast lightness correction. By analysing the imaging principle of the spots projected onto the surface of a spherical object, we regard the change in the position of the spots as a coded primitive. A binary-encoded M-array is designed by using primitives as the pattern of the NIR structured light. The stem-ends/calyxes can be identified by analysing a difference matrix from the NIR apple image captured with a multispectral camera. Fast lightness correction is performed to convert the uneven lightness distribution on the apple surface into a uniform lightness distribution over the whole fruit surface. The candidate defective regions segmented and extracted from the RGB apple image captured with the same multispectral camera are classified as the true defects or the stem-ends/calyxes by using the result of the stem-end/calyx identification in the NIR image. The apples are finally classified into sound and defective classes according to the existence or absence of defects respectively. The online experimental result with an average overall recognition accuracy of 90.2% for three apple varieties indicates that the proposed method is effective and suitable for defective apple detection.