Electronic nose (E-nose) plays an important role in the detection of Chinese baijiu, which is an alcoholic beverage of high reputation. However, traditional e-nose data processing relies on manual selection of features with complicated preprocessing steps. The production process of baijiu with the same flavor is similar, but due to the limited number of sensors, it is not easy to quickly extract specific features with e-nose. To overcome this shortcoming, we propose a novel method based on Channel Attention Convolutional Neural Network (CA-CNN) for authenticity identification of Chinese baijiu. The underlying channel attention module analyzes the dependencies between channels and learning weights in order to improve the detection accuracy. In particular, we evaluated the comprehensive performance of the system by comparing it with traditional machine learning methods. The prediction accuracy of the CA-CNN (98.53%) is better than the back-propagation artificial neural network (BP-ANN) (90.83%), the support vector machine (SVM) (84.5%), and the random forest (RF) (86.867%). The experiment results show that the CA-CNN has reasonable reliability, stability and good prediction performance in the quality classification of Chinese baijiu. It provides an effective reference method for the standardization of baijiu quality inspection.

Odor descriptors are words used to express human olfactory perception. At a certain level, predicting the odor descriptor rating using an electronic nose(E-nose) equips the machine with the ability to perceive odors. In this paper, we propose a novel deep learning framework for predicting odor descriptor rating via E-nose signals. The proposed framework has multiple sibling neural networks, including a Convolutional LSTM(ConvLSTM) layer and a regression layer. The ConvLSTM layers are used to learn the spatiotemporal features from sensor signal. On the one hand, it effectively extracts the temporal features of the signal, and on the other hand, it makes good use of the sensor dependencies. The output of the regression layer corresponds to a combination of descriptors or single descriptors, depending on their similarity. The experiment results show that: i) the framework utilizing the spatiotemporal correlation of the E-nose signal is more effective than using the temporal or spatial model alone, and that ii) the combination of descriptors based on descriptors similarity can effectively improve the prediction accuracy. On the DREAM dataset, our proposed framework outperforms the state-of-the- art methods in odor descriptor rating prediction.

The integrated electronic nose (e-nose) design, which integrates sensor arrays and recognition algorithms, has been widely used in different fields. However, the current integrated e-nose system usually suffers from the problem of low accuracy with simple algorithm structure and slow speed with complex algorithm structure. In this article, we propose a method for implementing a deep neural network for odor identification in a small-scale Field-Programmable Gate Array (FPGA). First, a lightweight odor identification with depthwise separable convolutional neural network (OI-DSCNN) is proposed to reduce parameters and accelerate hardware implementation performance. Next, the OI-DSCNN is implemented in a Zynq-7020 SoC chip based on the quantization method, namely, the saturation-flooring KL divergence scheme (SF-KL). The OI-DSCNN was conducted on the Chinese herbal medicine dataset, and simulation experiments and hardware implementation validate its effectiveness. These findings shed light on quick and accurate odor identification in the FPGA.

2D position information of input tokens is essential for transformer-based semantic segmentation models, especially on high-resolution aerial images. However, recent transformer-based segmentation methods use the position encoding to record position information and most position encoding methods encode the 1D positions of tokens. Therefore, we propose a two-dimensional semantic transformer model (2DSegFormer) for semantic segmentation on aerial images. In 2DSegFormer, we design a novel 2D positional attention to accurately record the 2D position information required by the transformer. Furthermore, we design the dilated residual connection and use it instead of skip connection in the deep stages to get a larger receptive field. skip connections are used in the shallow stages of 2DSegFormer to pass the details to the corresponding stages in the decoder. Experimental results on UAVid, Vaihingen and AeroScapes data sets demonstrate the effectiveness of 2DSegFormer. Compared with the state-of-the-art methods, 2DSegFormer shows better performance and great robustness on three different data sets. In particular, 2DSegFormer-B2 achieves first place in the public ranking on the UAVid test set.