A residual network with attention module for hyperspectral information of recognition to trace the origin of rice

Research on variety identification of common bean seeds based on hyperspectral and deep learning

Accurate, fast and non-destructive identification of varieties of common bean seeds is important for the cultivation and efficient utilization of common beans. This study is based on hyperspectral and deep learning to identify the varieties of common bean seeds non-destructively. In this study, the average spectrum of 3078 hyperspectral images from 500 varieties was obtained after image segmentation and sensitive region extraction, and the Synthetic Minority Over-sampling Technique (SMOTE) was used to achieve the equilibrium of the samples of various varieties. A one-dimensional convolutional neural network model (IResCNN) incorporating Inception module and residual structure was proposed to identify seed varieties, and Support Vector Machine (SVM), K-Nearest Neighbor (KNN), VGG19, AlexNet, ResNet50 were established to compare the identification effect. After analyzing the effects of multiple spectral preprocessing methods on the model, the study selected Savitzky-Golay smoothing correction (SG) for spectral preprocessing and extracted 66 characteristic wavelengths using Successive Projections Algorithm (SPA) as inputs to the discriminative model. Ultimately, the IResCNN model achieved the highest accuracy of 93.06 % on the test set, indicating that hyperspectral technology can accurately identify bean varieties, and the study provides a correct method of thinking for the non-destructive classification of multi-species small-sample bean varieties.

Non-destructive detection of defective maize kernels using hyperspectral imaging and convolutional neural network with attention module

Rapid, effective and non-destructive detection of the defective maize kernels is crucial for their high-quality storage in granary. Hyperspectral imaging (HSI) coupled with convolutional neural network (CNN) based on spectral and spatial attention (Spl-Spal-At) module was proposed for identifying the different types of maize kernels. The HSI data within 380-1000 nm of six classes of sprouted, heat-damaged, insect-damaged, moldy, broken and healthy kernels was collected. The CNN-Spl-At, CNN-Spal-At and CNN-Spl-Spal-At models were established based on the spectra, images and their fusion features as inputs for the recognition of different kernels. Further compared the performances of proposed models and conventional models were built by support vector machine (SVM) and extreme learning machine (ELM). The results indicated that the recognition ability of CNN with attention series models was significantly better than that of SVM and ELM models and fused features were more conducive to expressing the appearance of different kernels than single features. And the CNN-Spl-Spal-At model had an optimal recognition result with high average classification accuracy of 98.04 % and 94.56 % for the training and testing sets, respectively. The recognition results were visually presented on the surface image of kernels with different colors. The CNN-Spl-Spal-At model was built in this study could effectively detect defective maize kernels, and it also had great potential to provide the analysis approaches for the development of non-destructive testing equipment based on HSI technique for maize quality.

RPIOSL: construction of the radiation transfer model for rice leaves

The radiative transfer model of vegetation leaves simulates the transmission mechanism of light inside the vegetation and simulates the reflectivity of blades according to the change law of different components in the process of plant growth. Based on the PIOSL model, this paper combines PIOSL with the structure of rice leaves to construct a radiation transfer model for rice leaves. The parameters of each layer of the RPIOSL model are determined by the Non-dominated Sorting Genetic Algorithm-III. (NSGA-III.) algorithm. The reflectance spectra of 218 rice leaf samples in different periods were simulated using the RPIOSL model. The results show that the mean (RMSE) between the simulated and measured spectra of the constructed RPIOSL model is 0.1074, which is 0.0191 lower than that of the PROSPECT model. Among them, the spectral simulation effect of RPIOSL model in yellow and red light band is the best, and the RMSE at tillering period, jointing period, heading period and grouting period are 0.0584, 0.0576, 0.0724 and 0.0820, respectively. Therefore, the establishment of the RPIOSL model can accurately describe the interaction mechanism between light, which is of great significance for the rapid acquisition of rice growth information and accurate crop management.

Identification of white degradable and non-degradable plastics in food field: A dynamic residual network coupled with hyperspectral technology

In the food field, with the improvement of people's health and environmental protection awareness, degradable plastics have become a trend to replace non-degradable plastics. However, their appearance is very similar, making it difficult to distinguish them. This work proposed a rapid identification method for white non-degradable and degradable plastics. Firstly, a hyperspectral imaging system was used to collect the hyperspectral images of the plastics in visible and near-infrared bands (380-1038 nm). Secondly, a residual network (ResNet) was designed according to the characteristics of hyperspectral information. Finally, a dynamic convolution module was introduced into the ResNet to establish a dynamic residual network (Dy-ResNet) to adaptively mine the data features and realize the classification of the degradable and non-degradable plastics. Dy-ResNet had better classification performance than the other classical deep learning methods. The classification accuracy of the degradable and non-degradable plastics was 99.06%. In conclusion, hyperspectral imaging technology was combined with Dy-ResNet to identify the white non-degradable and degradable plastics effectively.

AUNet: a deep learning method for spectral information classification to identify inks

It is common to tamper with the contents of documents and forge contracts illegally. In this work, we propose a U-shaped network with attention modules (AUNet) and combine it with a hyperspectral system to effectively identify different inks. It provides an effective detection method for illegal tampering with documents and forging contract contents. First, the hyperspectral system obtains the spectral information of different pen inks without destroying the sample. Second, because the hyperspectral system's detection data have the characteristics of small samples, we introduce U-Net to conduct the deep fusion of multi-level spectral information to avoid feature degradation and fully mine the deep features hidden in the spectral information. Finally, spatial and channel attention modules are introduced to focus on the features affecting classification performance. The results show that AUNet effectively realizes the effective classification of ink spectral information and achieves 97.81% accuracy, 98.71% recall, 98.80% precision, and 98.71% F₁-score.

A multi-kernel channel attention combined with convolutional neural network to identify spectral information for tracing the origins of rice samples

Rice is a primary food consumed daily by many people, and different samples of rice often show disparate quality levels due to different production environments. In the rice market, it is common to sell low-quality rice with high-quality origin labels. As a nondestructive testing technology, spectral analysis has been widely used in food quality supervision. In this work, a deep learning method was developed and combined with a hyperspectral imaging system to achieve a quality-based identification of rice samples from different origins. First, the hyperspectral system was used to obtain spectral information of rice samples from five different origins. Then, a multi-kernel channel attention (MKCA) was proposed to focus on the deep features of the spectral information. Finally, based on the classical deep learning network, combined with MKCA, the spectral characteristics of rice samples from different origins were effectively identified. The results showed that MKCA combined with the LeNet-5 network structure achieved 97.40% accuracy, 97.63% precision, 97.78% recall, and 97.70% F₁-score. It provides an effective technical method for tracing rice.

Recent Advances in Counterfeit Art, Document, Photo, Hologram, and Currency Detection Using Hyperspectral Imaging

Forgery and tampering continue to provide unnecessary economic burdens. Although new anti-forgery and counterfeiting technologies arise, they inadvertently lead to the sophistication of forgery techniques over time, to a point where detection is no longer viable without technological aid. Among the various optical techniques, one of the recently used techniques to detect counterfeit products is HSI, which captures a range of electromagnetic data. To aid in the further exploration and eventual application of the technique, this study categorizes and summarizes existing related studies on hyperspectral imaging and creates a mini meta-analysis of this stream of literature. The literature review has been classified based on the product HSI has used in counterfeit documents, photos, holograms, artwork, and currency detection.

Estimation Model of Potassium Content in Cotton Leaves Based on Wavelet Decomposition Spectra and Image Combination Features

Potassium (K) is one of the most important elements influencing cotton metabolism, quality, and yield. Due to the characteristics of strong fluidity and fast redistribution of the K in plants, it leads to rapid transformation of the K lack or abundance in plant leaves; therefore, rapid and accurate estimation of potassium content in leaves (LKC, %) is a necessary prerequisite to solve the regulation of plant potassium. In this study, we concentrated on the LKC of cotton in different growth stages, an estimation model based on the combined characteristics of wavelet decomposition spectra and image was proposed, and discussed the potential of different combined features in accurate estimation of the LKC. We collected hyperspectral imaging data of 60 main-stem leaves at the budding, flowering, and boll setting stages of cotton, respectively. The original spectrum (R) is decomposed by continuous wavelet transform (CWT). The competitive adaptive reweighted sampling (CARS) and random frog (RF) algorithms combined with partial least squares regression (PLSR) model were used to determine the optimal decomposition scale and characteristic wavelengths at three growth stages. Based on the best "CWT spectra" model, the grayscale image databases were constructed, and the image features were extracted by using color moment and gray level co-occurrence matrix (GLCM). The results showed that the best decomposition scales of the three growth stages were CWT-1, 3, and 9. The best growth stage for estimating LKC in cotton was the boll setting stage, with the feature combination of "CWT-9 spectra + texture," and its determination coefficients (R ²val) and root mean squared error (RMSEval) values were 0.90 and 0.20. Compared with the single R model (R ²val = 0.66, RMSEval = 0.34), the R ²val increased by 0.24. Different from our hypothesis, the combined feature based on "CWT spectra + color + texture" cannot significantly improve the estimation accuracy of the model, it means that the performance of the estimation model established with more feature information is not correspondingly better. Moreover, the texture features contributed more to the improvement of model performance than color features did. These results provide a reference for rapid and non-destructive monitoring of the LKC in cotton.

Identification of Near Geographical Origin of Wolfberries by a Combination of Hyperspectral Imaging and Multi-Task Residual Fully Convolutional Network

Ningxia wolfberry is the only wolfberry product with medicinal value in China. However, the nutritional elements, active ingredients, and economic value of the wolfberry vary considerably among different origins in Ningxia. It is difficult to determine the origin of wolfberry by traditional methods due to the same variety, similar origins, and external characteristics. In the study, we have for the first time used a multi-task residual fully convolutional network (MRes-FCN) under Bayesian optimized architecture for imaging from visible-near-infrared (Vis-NIR, 400–1000 nm) and near-infrared (NIR-1700 nm) hyperspectral imaging (HSI) technology to establish a classification model for near geographic origin of Ningxia wolfberries (Zhongning, Guyuan, Tongxin, and Huinong). The denoising auto-encoder (DAE) was used to generate augmented data, then principal component analysis (PCA) was combined with gray level co-occurrence matrix (GLCM) to extract the texture features. Finally, three datasets (HSI, DAE, and texture) were added to the multi-task model. The reshaped data were up-sampled using transposed convolution. After data-sparse processing, the backbone network was imported to train the model. The results showed that the MRes-FCN model exhibited excellent performance, with the accuracies of the full spectrum and optimum characteristic spectrum of 95.54% and 96.43%, respectively. This study has demonstrated that the MRes-FCN model based on Bayesian optimization and DAE data augmentation strategy may be used to identify the near geographical origin of wolfberries.

Characterization of invisible symptoms caused by early phosphorus deficiency in cucumber plants using near-infrared hyperspectral imaging technology

In the early stage of P deficiency in cucumbers, the P deficiency symptoms in leaves are similar to the symptoms in control leaves at the early stage of aging and are difficult to identify with naked eyes or computer image processing techniques. In order to realize the quick diagnosis of P deficiency in plants at the early stage, the NIR hyperspectral images of control leaves and P-deficient leaves were collected, and the feature information of the NIR hyperspectral images was extracted by PCA and ICA respectively. Through PCA and HCA verification, the IC1 component diagram of P-deficient leaves NIR hyperspectral image could effectively characterize the features of invisible water-stained plaques caused by early P-deficient leaves. Region of interest from IC1 was selected to extract spectral information for classification, and the diagnostic rate was remarkably improved. Finally, 240 leaves were diagnosed by using the BP-ANN model with a diagnostic rate of 97.5%. In addition, the experiment verified that it was possible to diagnose whether the plant was in the state of P deficiency 21 days in advance, and timely guidance of top dressing was of great significance to increase yield.