A procedure for the prediction of talc content in wheat flour based on radial basis function (RBF) neural network and near-infrared spectroscopy (NIRS) data is described. In this study, 41 wheat flour samples adulterated with different concentrations of talc were used. The diffuse reflectance spectra of all samples were collected by NIRS analyzer in the spectral range of 400 to 2,500 nm. A sample of outliers was eliminated by Mahalanobis distance based on near-infrared spectral scanning, and the remaining 40 wheat flour samples were used for spectral characteristic analysis. A calibration set of 26 samples and a prediction set of 14 samples of wheat flour were built as a result of sample set partitioning based on joint x-y distances division. A comparison of Savitzky-Golay smoothing, multiplicative scatter correction (MSC), first derivation, second derivation, and standard normal variation in the modeling showed that MSC has the best preprocessing effect. To develop a simpler, more efficient prediction model, the correlation coefficient method (CCM) was used to reduce spectral redundancy and determine the maximum correlation informative wavelength (MIW). From the full 1,050 wavelengths, 59 individual MIWs were finally selected. The optimal combined detection model was CCM-MSC-RBF based on the selected MIWs, with a determination of prediction coefficients of prediction (Rp) of 0.9999, root-mean-square error of prediction of 0.0765, and residual predictive deviation of 65.0909. The study serves as a proof of concept that NIRS technology combined with multivariate analysis has the potential to provide a fast, nondestructive and reliable assay for the prediction of talc content in wheat flour.

Taking a variety of edible oils as the research object, including soybean oil, peanut oil, rapeseed oil, a method based on Near-Infrared Spectroscopy (NIRS) to identify the frying times is proposed to evaluate the quality of frying oil. Ten rounds of frying experiments are carried out for each of the three oils. The spectra of the first eight rounds are used to build the model, and the last two are used for model testing. First, all the original spectra are preprocessed using the first derivative (1D). Then, the correlation coefficient between the sequence of frying times and absorbance is calculated, and the characteristic wavelengths with a high correlation coefficient are extracted. Finally, a differential prediction model is established based on the characteristic wavelengths. The results show that the differential prediction model accurately predicts the frying times of various edible oils and provides a new method for quality inspection of frying oil, and the predicted accuracy of the frying times of three frying oils is 100% within the allowable range of error.

Pulsar observation and research are of great significance. With the gradual increase in the performance and quantity of observing equipment, the received pulsar observation data also increase geometrically. Machine learning can mine large-scale pulsar data sets to realize automatic pulsar candidate identification. However, the pulsar candidate and non-pulsar candidate data sets are highly imbalanced. The traditional classifier trained on the data set with imbalanced class distribution usually shows poor generalization performance, which will cause the classifier to be biased towards negative instances. To address the problem of poor identification algorithm performance caused by sample imbalance, we describe a new framework, namely AdaBoost-multi-input-CNN (AdaBoost-MICNN). In AdaBoost, multiple baseline models are trained sequentially, and the weight of each training sample changes as the training progresses. Especially the minority samples, which are more easily ignored, will get more attention in subsequent baseline models. This will solve the problem of imbalanced classification. Meanwhile, there are usually four pulsar diagnostic plots. To automatically extract and integrate these four diagnostic plots features, we use multi-input-CNN (MICNN) as the baseline model of AdaBoost. The attention mechanism is introduced to enhance the performance of MICNN’s extraction of features, and design an objective function that adds the maximum mean discrepancy to enhance the anti-interference ability of the model. Finally, the proposed method is tested in the High Time Resolution Universe Medlat Data, and the experimental results verify the effectiveness and efficiency of the method.

With the development of radio telescope technology, the quantity and types of acquired pulsar candidate data have increased dramatically. However, it is difficult to accurately identify pulsar candidates. Therefore, we propose to use multimodal fusion technology, called the multimodal fusion-based pulsar identification model (MFPIM), to build a deep learning model to improve the efficiency and accuracy of pulsar candidate identification. MFPIM treats each diagnostic plot of pulsar candidates as a modality and uses multiple convolutional neural networks to extract effective features from the diagnostic plots. After fusing the features, the commonality of different modalities in high-dimensional space is obtained to ensure that the model can take full advantage of the complementarity between diagnostic plots and thus identify pulsar candidates, achieved better classification performance than other current supervised learning algorithms. In addition, a channel attention mechanism is used in the model to enable it to learn the importance of different channel features so that the model focuses more on the channel information in the input data that is more meaningful for classification, reducing the model size while extracting pulsar diagnostic map features more accurately. We conducted experiments on the Five-hundred-meter Aperture Spherical radio Telescope (FAST) data set, and the results show that MFPIM can effectively identify the pulsars in the FAST data set with an identification accuracy of over 98%. To further verify the robustness of the model, we applied the MFPIM to the High Time Resolution Universe data set using transfer learning, with the test accuracy and F1 score reaching over 99%.