The attention mechanism enables graph neural networks (GNNs) to learn the attention weights between the target node and its one-hop neighbors, thereby improving the performance further. However, most existing GNNs are oriented toward homogeneous graphs, and in which each layer can only aggregate the information of one-hop neighbors. Stacking multilayer networks introduces considerable noise and easily leads to over smoothing. We propose here a multihop heterogeneous neighborhood information fusion graph representation learning method (MHNF). Specifically, we propose a hybrid metapath autonomous extraction model to efficiently extract multihop hybrid neighbors. Then, we formulate a hop-level heterogeneous information aggregation model, which selectively aggregates different-hop neighborhood information within the same hybrid metapath. Finally, a hierarchical semantic attention fusion model (HSAF) is constructed, which can efficiently integrate different-hop and different-path neighborhood information. In this fashion, this paper solves the problem of aggregating multihop neighborhood information and learning hybrid metapaths for target tasks. This mitigates the limitation of manually specifying metapaths. In addition, HSAF can extract the internal node information of the metapaths and better integrate the semantic information present at different levels. Experimental results on real datasets show that MHNF achieves the best or competitive performance against state-of-the-art baselines with only a fraction of 1/10 $\sim$ 1/100 parameters and computational budgets. Our code is publicly available at https://github.com/PHD-lanyu/MHNF

In this paper, we proposed a novel method to determine the oil film thickness on water with a wide range, which is based on the intensity ratio between the laser-induced oil film fluorescence and Raman scattering of the water without oil films. To verify the feasibility of this method, we presented a portable detection system which employs a laser diode at 405 nm and an intensified charged-coupled device (ICCD) detector. Using this system, the diesel oil films with variable thicknesses on water were measured at different distances. The measured thickness was in the range from 0.07 mm to 36.62 mm with an average relative error of 5.98% and 4.07% by parameter calibration method (PCM) and spectral fitting method (SFM), respectively. The experimental results showed that the effective thickness range measured using the proposed ratio method is 40 times greater than that measured by the traditional Raman suppression method (less than 0.82 mm). In addition, different thicknesses of crude oil films were also measured, and the effective thickness range was from 0.002 mm to 0.493 mm, which was 70 times greater than the traditional Raman method (less than 0.007 mm). In the range of 0.007 mm to 0.493 mm, the average relative error is less than 10% using the proposed ratio method. Furthermore, this method can greatly eliminate the influences from the detection distances, instrument factors and external environments compared with the fluorometric method.

Counterfeit Chinese liquor incidents in China have significantly disrupted market order and jeopardized the health of consumers. Currently, deep learning-based spectral detection techniques are extensively leveraged in non-invasive food inspection. Excessive reliance on labels severely limits its application in real scenarios. To make better use of limited samples, we are the first to use the “unsupervised pre-training & supervised fine-tuning” paradigm in combining the Transformer architecture for feature extraction and classification of the Chinese liquor spectrum, and propose Spectrum-BERT, which represents Bidirectional Encoder Representations from Transformers for Spectrum. Specifically, we creatively propose spectral curve partitioning and 1-D convolutional layer mapping to maintain the model’s sensitivity to characteristic peak locations and local information of spectral curves. Moreover, the paradigm of “unsupervised pre-training & supervised fine-tuning” addresses the limitation of label deficiency, thus improving the model’s applicability. Finally, we have conducted extensive experiments on the real liquor spectral dataset. Comparative experiments demonstrate that Spectrum-BERT outperforms all baselines on all metrics using only 70% supervised signal. The limit experimental results show that Spectrum-BERT can still maintain its lead using only the 10% supervised signal. Thanks to the more efficient model architecture, Spectrum-BERT’s model parameters and FLOPs are only 1/568 and 1/322 of those of baselines.

Laser-induced fluorescence lidar (LIF-LiDAR) is an effective technology for the stand-off detection, identification, and quantification of oil spills on the water surface. In this work, by using a self-developed LIF-LiDAR system, the laser-induced spectra of different oils with different thicknesses on the water surface were remotely measured at a distance of 100 m. The traditional algorithms based on the similarity degree of matching spectra are adopted to identify the oil types, which results in the 65% (min = 33.68%) and 85% (max = 98.66%) average matching accuracy when the pure oil spectra (infinite thickness) and the average oil spectra (different thickness) are introduced as the database, respectively. Furthermore, to avoid the influence of the pretreatment process in the traditional matching method, the machine learning model is applied to classify the oil types and an identifying accuracy of nearly 100% is successfully achieved. Our results presented in this work not only demonstrate the good performance of the classification models of Fine Tree, Linear Discriminant, Support Vector Machine, k-Nearest Neighbor, and Neural Network in remote oil type identification, but also the capability of the combination of LIF-LiDAR and machine learning algorithms on the high accuracy identification in the types of oil films on water.

In this invited paper, a novel water chemical oxygen demand (COD) detection method based on laser spectroscopy is proposed, and the COD value is obtained according to the intensity ratio of laser Fluorescence-Raman signal produced by laser incident into water. Based on the laser Fluorescence-Raman ratio (LFRR) method, we design a portable water quality measurement system. In which, a 405 nm semiconductor laser is used as the light source. The laser is coupled into the delivering port of Y-type fiber and then transmitted into water. The generated Mie scattered light, water Raman light and corresponding fluorescence are received and transmitted into spectrometer through the detecting port of Y-type fiber, and the COD value can be obtained by analyzing the LFRR of water spectra. The mixed solution of sodium humate and glucose is used as the COD standard solution to calibrate the designed system. The experiment results show that the optimal volume ratio of the two solutions is 1:29. When the COD of standard solutions is in the 1–12 mg/L concentration range, a good linear relationship can be found between the COD value and LFRR value with a correlation coefficient of 0.974. In addition, the COD of natural water samples are measured with LFRR method, the results of which are consistent with COD values obtained by rapid digestion spectrophotometry. Meanwhile, experimental results prove that the COD detection method proposed in this paper has the advantages of high sensitivity, high precision, high detection speed and simplicity, which can be widely used in various water areas for real-time COD monitoring.