The identification of pepper leaf diseases is crucial for ensuring the safety and quality of pepper yield. However, existing methods heavily rely on manual diagnosis, resulting in inefficiencies and inaccuracies. In this study, we propose a lightweight convolutional neural network (CNN) model for recognizing pepper leaf diseases and subsequently develop an application based on this model. To begin with, we acquired various images depicting healthy leaves as well as leaves affected by viral diseases, brown spots, and leaf mold. It is noteworthy that these images were captured against a background of human palms, which is commonly encountered in field conditions. The proposed CNN model adopts the GGM-VGG16 architecture, incorporating Ghost modules, global average pooling, and multi-scale convolution. Following training with the collected image dataset, the model was deployed on a mobile terminal, where an application for pepper leaf disease recognition was developed using Android Studio. Experimental results indicate that the proposed model achieved 100 % accuracy on images with a human palm background, while also demonstrating satisfactory performance on images with other backgrounds, achieving an accuracy of 87.38 %. Furthermore, the developed application has a compact size of only 12.84 MB and exhibits robust performance in recognizing pepper leaf diseases.

The segmentation of pepper leaves from pepper images is of great significance for the accurate control of pepper leaf diseases. To address the issue, we propose a bidirectional attention fusion network combing the convolution neural network (CNN) and Swin Transformer, called BAF-Net, to segment the pepper leaf image. Specially, BAF-Net first uses a multi-scale fusion feature (MSFF) branch to extract the long-range dependencies by constructing the cascaded Swin Transformer-based and CNN-based block, which is based on the U-shape architecture. Then, it uses a full-scale feature fusion (FSFF) branch to enhance the boundary information and attain the detailed information. Finally, an adaptive bidirectional attention module is designed to bridge the relation of the MSFF and FSFF features. The results on four pepper leaf datasets demonstrated that our model obtains F1 scores of 96.75%, 91.10%, 97.34% and 94.42%, and IoU of 95.68%, 86.76%, 96.12% and 91.44%, respectively. Compared to the state-of-the-art models, the proposed model achieves better segmentation performance. The code will be available at the website: https://github.com/fangchj2002/BAF-Net.

Introduction: Pepper leaf segmentation plays a pivotal role in monitoring pepper leaf diseases across diverse backgrounds and ensuring healthy pepper growth. However, existing Transformer-based segmentation methods grapple with computational inefficiency, excessive parameterization, and inadequate utilization of edge information.Methods: To address these challenges, this study introduces an Adaptive Multi-Scale MLP (AMS-MLP) framework. This framework integrates the Multi-Path Aggregation Module (MPAM) and the Multi-Scale Context Relation Mask Module (MCRD) to refine object boundaries in pepper leaf segmentation. The AMS-MLP includes an encoder, an Adaptive Multi-Scale MLP (AM-MLP) module, and a decoder. The encoder’s MPAM fuses five-scale features for accurate boundary extraction. The AM-MLP splits features into global and local branches, with an adaptive attention mechanism balancing them. The decoder enhances boundary feature extraction using MCRD.Results: To validate the proposed method, extensive experiments were conducted on three pepper leaf datasets with varying backgrounds. Results demonstrate mean Intersection over Union (mIoU) scores of 97.39%, 96.91%, and 97.91%, and F1 scores of 98.29%, 97.86%, and 98.51% across the datasets, respectively.Discussion: Comparative analysis with U-Net and state-of-the-art models reveals that the proposed method significantly improves the accuracy and efficiency of pepper leaf image segmentation.

Diesel-natural gas dual fuel engine has gained increasing interesting in recent years because of its excellent power and economy. However, the reliability of the dual fuel engine does not meet the requirements of practical application. The piston maximun temperature (PMT) of the dual fuel engine easily exceeds the security border. In view of this, this paper proposes a method based on the lasso regression to estimate the PMT of the dual fuel engine, so as to real-timely monitor the health state of the dual fuel engine. Specifically, PMTs under some working conditions were offline acquired by the finite element analysis with ANSYS. A model is presented to describe the relationship between the PMT and some indirect engine variables, including NOx emission, excess air coefficient, engine speed, and inlet pressure, and the model parameters are optimized using the lasso regression algorithm, which can be easily implemented by the electronic control unit (ECU). Finally, the model is employed to real-timely estimate the PMT of the dual fuel engine. Experiments reveal that the proposed model produces satisfying predictions with deviations less than 10 °C.

Semi-supervised learning (SSL) has garnered considerable attention in medical image segmentation due to its ability to leverage abundant unlabeled data, thereby significantly alleviating the dependency on expensive pixel-wise annotations. However, existing Mean-Teacher-based frameworks typically guide the student network using a teacher model in a unidirectional or loosely interactive manner, which fails to fully exploit the complementary relationships between different networks and often struggles with low-confidence predictions that degrade boundary accuracy. To address these limitations, we propose SDBS-Net, a novel semi-supervised dual-student self-distillation network for 3D medical image segmentation. The architecture comprises a shared encoder and two parallel decoders, forming two networks (Model I and Model II) that process both labeled and unlabeled volumes simultaneously. To enable effective bidirectional knowledge transfer, we introduce two synergistic training schemas: a prior knowledge learning (PKL) schema with hierarchical difference (HD) modules and a Self-Distillation Learning (SDL) schema. The PKL schema extracts and inject multi-scale discrepancy features from Model I into Model II, enhancing discriminative representation and boundary delineation, while the SDL schema employs multi-level soft-label and feature distillation losses to correct confirmation bias and refine overlooked regions in a mutually reinforcing manner. Extensive experiments on two challenging benchmarks such as the Left Atrium (LA) dataset and the Pancreas-CT dataset demonstrate the superiority of SDBS-Net. With only 10% and 20% labeled data, our method achieves Dice scores of 88.19% and 90.51% on the LA dataset, and 71.39% and 79.37% on the Pancreas-CT dataset, respectively, consistently outperforming state-of-the-art semi-supervised approaches while approaching fully supervised performance.