Search

Fruit quality assessment plays a crucial role in determining their market value, consumer acceptance, and post-harvest management. In recent years, spectroscopic techniques have gained significant attention as non-destructive methods for evaluating fruit quality. In this study, we propose a novel deep-learning network, called GCNN-LSTM-AT, for the prediction of five important parameters of tangerines using visible and near-infrared spectroscopy (Vis–NIR). The quality attributes include soluble solid content (SSC), total acidity (TA), acid–sugar ratio (A/S), firmness, and Vitamin C (VC). The proposed model combines the strengths of graph convolutional network (GCN), convolutional neural networks (CNNs), and long short-term memory (LSTM) to capture both spatial and sequential dependencies in the spectra data, and incorporates an attention mechanism to enhance the discriminative ability of the model. To investigate the effectiveness and stability of the model, comparisons with three traditional machine-learning algorithms—moving window partial least squares (MWPLS), random forest (RF), and support vector regression (SVR)—and two deep neural networks—DeepSpectra2D and CNN-AT—are provided. The results have shown that the GCNN-LSTM-AT network outperforms other algorithms and models, achieving accurate predictions for SSC (R2: 0.9885, RMSECV: 0.1430 ∘Brix), TA (R2: 0.8075, RMSECV: 0.0868%), A/S (R2: 0.9014, RMSECV: 1.9984), firmness (R2: 0.9472, RMSECV: 0.0294 kg), and VC (R2: 0.7386, RMSECV: 29.4104 mg/100 g) of tangerines.

Localization and tracking in multi-player sports present significant challenges, particularly in wide and crowded scenes where severe occlusions can occur. Traditional solutions relying on a single camera are limited in their ability to accurately identify players and may result in ambiguous detection. To overcome these challenges, we proposed fusing information from multiple cameras positioned around the field to improve positioning accuracy and eliminate occlusion effects. Specifically, we focused on soccer, a popular and representative multi-player sport, and developed a multi-view recording system based on a 1+N strategy. This system enabled us to construct a new benchmark dataset and continuously collect data from several sports fields. The dataset includes 17 sets of densely annotated multi-view videos, each lasting 2 min, as well as 1100+ min multi-view videos. It encompasses a wide range of game types and nearly all scenarios that could arise during real game tracking. Finally, we conducted a thorough assessment of four multi-view multi-object tracking (MVMOT) methods and gained valuable insights into the tracking process in actual games.

Multi-player action recognition for automatic analysis in sports is the subject of increasing attention. Trajectory-tracking technology is key for accurate recognition, but little research has focused on this aspect, especially for non-professional matches. Here, we study multi-player tracking in the most popular and complex sport among non-professionals—soccer. In this non-professional soccer player tracking (NPSPT) challenge, single-view-based motion recording systems for continuous data collection were installed in several soccer fields, and a new benchmark dataset was collected. The dataset consists of 17 2-min long super-high-resolution videos with diverse game types consistently labeled across time, covering almost all possible situations for multi-player detection and tracking in real games. A comprehensive evaluation was conducted on the state-of-the-art multi-object-Tracking (MOT) systems, revealing insights into player tracking in real games. Our challenge introduces a new dimension for researchers in the player recognition field and will be beneficial to further studies.

DeepLab v3+ neural network shows excellent performance in semantic segmentation. In this paper, we proposed a segmentation framework based on DeepLab v3+ neural network and applied it to the problem of hyperspectral imagery classification (HSIC). The dimensionality reduction of the hyperspectral image is performed using principal component analysis (PCA). DeepLab v3+ is used to extract spatial features, and those are fused with spectral features. A support vector machine (SVM) classifier is used for fitting and classification. Experimental results show that the framework proposed in this paper outperforms most traditional machine learning algorithms and deep-learning algorithms in hyperspectral imagery classification tasks.

In this paper, we show that the guided mode resonance can exist in a low-index waveguide layer on top of a high-index substrate. With the help of the interaction of diffraction from a metal grating and total internal reflection effects, we verify that the guided mode can be supported in the low-index SU8 layer on a high-index substrate. Simulation and experiment show the resonant wavelength can be simply manipulated by controlling the geometrical parameters of the metal grating and waveguide layer. This structure extends the possibilities of guided-mode resonance to a broader class of functional materials and may boost its use in applications such as field enhancement, sensing and display.