Jinan Gu’s research while affiliated with Jiangsu University and other places

With the continuous development of artificial intelligence technology, the transformation of traditional agriculture into intelligent agriculture is quickly accelerating. However, due to the diverse growth postures of tender shoots and complex growth environments in tea plants, traditional tea picking machines are unable to precisely select the tender shoots, and the picking of high-end and premium tea still relies on manual labor, resulting in low efficiency and high costs. To address these issues, an instance segmentation algorithm named YOLOv8-TEA is proposed. Firstly, this algorithm is based on the single-stage instance segmentation algorithm YOLOv8-seg, replacing some C2f modules in the original feature extraction network with MVB, combining the advantages of convolutional neural networks (CNN) and Transformers, and adding a C2PSA module following spatial pyramid pooling (SPPF) to integrate convolution and attention mechanisms. Secondly, a learnable dynamic upsampling method is used to replace the traditional upsampling, and the CoTAttention module is added, along with the fusion of dilated convolutions in the segmentation head to enhance the learning ability of the feature fusion network. Finally, through ablation experiments and comparative experiments, the improved algorithm significantly improves the segmentation accuracy while effectively reducing the model parameters, with mAP (Box) and mAP (Mask) reaching 86.9% and 86.8%, respectively, and GFLOPs reduced to 52.7.

In the context of Industry 4.0, new sensing and communication technologies have unlocked vast amounts of process data, offering significant potential for its transformation into actionable insights to support manufacturing decisions. The reliable detection and diagnosis of faults in rolling element bearings pose a significant challenge for condition-based maintenance and fault detection and diagnosis (FDD), which are critical strategies for enhancing equipment reliability and reducing operational costs. Deep learning methods, such as convolutional neural networks (CNNs), can extract features from vibration signals compared to traditional signal processing. However, these methods in isolation are insufficient to reliably detect novel fault conditions and faults in variable working environments. Also, existing novelty and anomaly detection criteria are not accurate enough to correctly distinguish novel or unseen faults. This study introduces a multi-fault detection framework leveraging a variational autoencoder with Mahalanobis distance (MD) novelty scores for unknown condition detection and a hybrid CNN-Swin transformer (Swin-T) model for incremental learning and fault classification. Using frequency-domain transformation and image-based representation of vibration signals, a hybrid model with a CNN-based feature extractor after projecting to the patch embedding layer of a simplified Swin-T model is trained incrementally with novel conditions to allow continuous learning and adaptation. Extensive validation with three separate datasets from fault simulation test rigs demonstrates the superior performance of the method over traditional and cutting-edge models in FDD and novelty detection (ND), achieving near-perfect accuracy (99.7%), precision (99.8%), recall (99.6%), and F1 score (99.7%). ND outperformed traditional approaches with an MD novelty score threshold yielding a true-positive rate of 98.9% and a false-positive rate of 1.2%. Additionally, incremental learning improved classification accuracy by up to 5.4% for newly introduced fault types, highlighting its adaptability. These results demonstrate the framework’s ability to enhance reliability and efficiency in industrial machinery maintenance by identifying both known and novel fault conditions with high precision.

The detection of defects on industrial surfaces is essential for guaranteeing the quality and safety of products. Deep learning‐based object detection methods have demonstrated impressive efficacy in industrial applications in recent years. However, due to the complex and variable shape of defects, the similarity between defects and background, large intra‐class differences, and small inter‐class differences lead to low classification accuracy, it is a great challenge to achieve accurate defect detection. To overcome these challenges, this research proposed a novel network specifically designed for defect detection. First, a feature extraction network, ResDCA‐Net, is constructed based on deformable convolution and lightweight multi‐scale attention, where deformable convolution can adaptively adjust to extract features of defects with complex and variable shapes. Second, the lightweight multi‐scale attention module is constructed, which uses multi‐branch and cross‐space fusion to obtain the complete feature space attention map, thereby improving the defect feature attention and reducing the background feature attention. Third, to enhance the classification and localization accuracy, an attention‐based decoupled prediction module is proposed to ensure that the classification and regression branches of the model can focus on their required features. Finally, extensive comparative experiments indicate that the proposed approach performs best, achieving 83.7% and 83.4% mean Average Precision (mAP) on the GC10‐DET and NEU‐DET datasets, respectively. The effectiveness of the proposed individual modules is further validated in ablation experiments, which demonstrate the excellent performance and potential in defect detection tasks.

Intelligent apple-picking robots can significantly improve the efficiency of apple picking, and the realization of fast and accurate recognition and localization of apples is the prerequisite and foundation for the operation of picking robots. Existing apple recognition and localization methods primarily focus on object detection and semantic segmentation techniques. However, these methods often suffer from localization errors when facing occlusion and overlapping issues. Furthermore, the few instance segmentation methods are also inefficient and heavily dependent on detection results. Therefore, this paper proposes an apple recognition and localization method based on RGB-D and an improved SOLOv2 instance segmentation approach. To improve the efficiency of the instance segmentation network, the EfficientNetV2 is employed as the feature extraction network, known for its high parameter efficiency. To enhance segmentation accuracy when apples are occluded or overlapping, a lightweight spatial attention module is proposed. This module improves the model position sensitivity so that positional features can differentiate between overlapping objects when their semantic features are similar. To accurately determine the apple-picking points, an RGB-D-based apple localization method is introduced. Through comparative experimental analysis, the improved SOLOv2 instance segmentation method has demonstrated remarkable performance. Compared to SOLOv2, the F1 score, mAP, and mIoU on the apple instance segmentation dataset have increased by 2.4, 3.6, and 3.8%, respectively. Additionally, the model’s Params and FLOPs have decreased by 1.94M and 31 GFLOPs, respectively. A total of 60 samples were gathered for the analysis of localization errors. The findings indicate that the proposed method achieves high precision in localization, with errors in the X, Y, and Z axes ranging from 0 to 3.95 mm, 0 to 5.16 mm, and 0 to 1 mm, respectively.