The feeding rate is a critical reference parameter for the efficient and low-loss operation control of grain combine harvesters. Current feeding rate measurement technologies face challenges such as low accuracy, poor stability, and delayed measurement results. Rapidly and accurately obtaining the feeding rate during the operation of combine harvesters has become a key focus and challenge in the field of agricultural machinery research. This paper proposes a feeding rate measurement method based on multimodal information fusion, including crop visual perception and harvester cutting load. This method first uses an image sensor to capture crop images in the working area in front of the combine harvester and proposes an improved Deeplab image segmentation algorithm to achieve accurate crop height measurement. Then, the real-time load information of the combine harvester's cutter is used to calculate the cutting area. Finally, by combining the operational speed information of the harvester, the precise feeding volume is calculated, effectively addressing the interference caused by complex working conditions in feeding rate measurement. To reduce the interference of combine harvester vibrations and electromagnetic environments on sensor signals, an M-E fusion filter is proposed to extract effective cutting resistance information, achieving accurate and stable calculation of header resistance information. Using the ROS communication mechanism combined with the Python 3.6 programming language, an intelligent measurement system software for the feeding rate of grain combine harvesters was developed. This system realizes functions such as sensor data acquisition, multi-sensor data fusion computation, and human-computer interaction. The results of the field tests showed that at various operating speeds of 2 km/h, 3 km/h, 4 km/h, and 5 km/h, the average relative errors of the feeding rate measurement results are 1.46%, 1.77%, 1.76%, and 1.53%, respectively, with an average relative error of 1.64% and a root mean square error average of 0.103×10⁻³ m³·s⁻¹. At various crop densities of 553 plants/m², 415 plants/m², 277 plants/m², and 138 plants/m², the average relative errors of the feeding rate measurement results are 1.77%, 1.5%, 2.39%, and 1.91%, respectively, with an average relative error of 1.89% and a root mean square error average of 0.061×10⁻³ m³·s⁻¹. The feeding rate measurement system developed in this study has high accuracy and strong stability, providing important measurement technology support for intelligent control research of combine harvesters and has high practical value.

Conventional texture instruments rely on limited feature parameters, which poses challenges for the accurate quantitative evaluation of food graininess. In this study, a multidimensional tactile feature fusion method was proposed to assess food graininess, by using a self-developed tactile information acquisition system. A total of 180 sets of tactile friction and vibration data were collected based on six types of food-saliva mixture samples. Four machine learning algorithms were applied for feature extraction, and the convolutional neural network (CNN) exhibited the best performance, enabling the identification of nine key tactile friction and vibration features. These features, together with sensory evaluation scores, were utilized for correlation analysis, qualitative discrimination, and quantitative prediction. The results demonstrated that tactile vibration features exhibited a stronger correlation with sensory graininess than friction features. Principal component analysis (PCA) effectively discriminated food samples with different graininess levels, achieving an accuracy exceeding 95.8%. Furthermore, graininess was accurately predicted by a stepwise multiple linear regression (SMLR) model (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$R_{p} = 0.889$</tex-math></inline-formula>, RMSEP = 0.622). Overall, these findings confirm the effectiveness of multidimensional tactile feature information fusion methods for evaluating food graininess and highlight their potential for the quantitative characterization of food texture.

Accurate classification of hyperspectral images (HSI) is crucial for earth observation and agricultural production analysis, yet remains challenging due to high dimensionality, spectral variability, and limited training samples. Traditional approaches often struggle to effectively balance computational complexity with the ability to capture spatial-spectral relationships of observation targets. To address this challenge, we propose OmniMamba, a novel omnidirectional state space model that adopts a collaborative alternation strategy integrating single-scale and multi-scale feature processing with omnidirectional scanning mechanisms. Four complementary scanning patterns (row, column, zigzag, and snake) are employed in the omnidirectional scanning mechanism to transfer 2D spatial data into 1D spatially structured feature sequence, which preserves directional sensitivity while achieving global dependency modeling with only linear complexity. This avoids the quadratic complexity bottleneck inherent in self-attention mechanisms. Our collaborative alternation strategy coordinates fine-grained spectral signatures with hierarchical spatial contexts through cascaded processing stages, addressing the spectral-spatial feature fusion challenge in HSI classification. Extensive experiments conducted on four benchmark datasets validate the superiority of OmniMamba, achieving a mean overall accuracy of 99.28%, significantly outperforming existing methods. Remarkably, our model accomplishes the performance with only 246 K parameters and 0.04 GFLOPs, demonstrating dramatically low computational complexity than the state-of-the art conventional CNN and transformer-based architectures.

The recent surge in open-source Multimodal Large Language Models (MLLM) frameworks, such as LLaVA, provides a convenient kickoff for artificial intelligence developers and researchers. However, most of the MLLM frameworks take vision as the main input modality, and provide limited in-depth support for the modality of speech, audio, and music. This situation hinders the development of audio-language models, and forces researchers to spend a lot of effort on code writing and hyperparameter tuning. We present SLAM-LLM, an open-source deep learning framework designed to train customized MLLMs, focused on speech, language, audio, and music processing. SLAM-LLM provides a modular configuration of different encoders, projectors, LLMs, and parameter-efficient fine-tuning plugins. SLAM-LLM also includes detailed training and inference recipes for mainstream tasks, along with high-performance checkpoints like LLM-based Automatic Speech Recognition (ASR), Automated Audio Captioning (AAC), and Music Captioning (MC). Some of these recipes have already reached or are nearing state-of-the-art performance, and some relevant techniques have also been accepted by academic papers. We hope SLAM-LLM will accelerate iteration, development, data engineering, and model training for researchers. We are committed to continually pushing forward audio-based MLLMs through this open-source framework, and call on the community to contribute to the LLM-based speech, audio and music processing.

Tractography fiber clustering using diffusion MRI (dMRI) is a crucial method for white matter (WM) parcellation to enable analysis of brain's structural connectivity in health and disease. Current fiber clustering strategies primarily use the fiber geometric characteristics (i.e., the spatial trajectories) to group similar fibers into clusters, while neglecting the functional and microstructural information of the fiber tracts. There is increasing evidence that neural activity in the WM can be measured using functional MRI (fMRI), providing potentially valuable multimodal information for fiber clustering to enhance its functional coherence. Furthermore, microstructural features such as fractional anisotropy (FA) can be computed from dMRI as additional information to ensure the anatomical coherence of the clusters. In this paper, we develop a novel deep learning fiber clustering framework, namely <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Deep Multi-view Fiber Clustering (DMVFC)</i>, which uses joint multi-modal dMRI and fMRI data to enable functionally consistent WM parcellation. DMVFC can effectively integrate the geometric and microstructural characteristics of the WM fibers with the fMRI BOLD signals along the fiber tracts. DMVFC includes two major components: (1) a multi-view pretraining module to compute embedding features from each source of information separately, including fiber geometry, microstructure measures, and functional signals, and (2) a collaborative fine-tuning module to simultaneously refine the differences of embeddings. In the experiments, we compare DMVFC with two state-of-the-art fiber clustering methods and demonstrate superior performance in achieving functionally meaningful and consistent WM parcellation results.

Due to the sparsity of scattered signals, the theory of compressed sensing (CS) has natural applicability in the field of millimetre-wave (MMW) near-field synthetic aperture radar (SAR) imaging. However, the sparse basis matrix in conventional sparse imaging model not only has a high demand for storage resources, but also destroys the block-sparse structure of the SAR image, which leads to the difficulties in applying CS based imaging algorithms to the practical system and hinders relevant theoretical research to some extent. To overcome these problems, a sparse imaging model based on the discrete Fourier transform basis is proposed, which significantly reduces the size of sensing matrix and improves its generation speed while preserving the block-sparse structure, allowing better-performing block-sparse recovery algorithms to be employed for imaging. In order to design the corresponding sparse imaging algorithm and provide a reference for others, the generalized approximate message passing based pattern-coupled sparse Bayesian learning (GAMP-PCSBL), which can achieve satisfactory sparse recovery results through neighborhood association and approximate computation of the posterior distribution, is introduced with some modifications. Additionally, a generalized incremental 3D imaging framework based on the model is proposed, enabling efficient 3D imaging. Experiments based on public data sets demonstrate the effectiveness of proposed imaging model, algorithms and framework. Only approximately 6<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathbf{\%}$</tex-math></inline-formula> and 14<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathbf{\%}$</tex-math></inline-formula> of the entire data are required for single-target and multi-target imaging respectively.