1 January 1999 in Measurement Science and Technology
1 January 2023 in IEEE Access
The importance of object detection in intelligent logistics applications is increasingly recognized. However, current detector models suffer from challenges such as high computational cost and low detection accuracy, which limit their deployment on edge devices with limited computational power in logistics scenarios. To address these issues, this paper proposes a novel lightweight detector model (GBForkDet) based on YOLOv8 for forklift safety driving. Firstly, the Ghost module is integrated into YOLOv8 to optimize the Backbone feature extraction process, reducing the computational cost of the model. Then, a Bi-directional Omni-Dimensional Dynamic Neck (BiODNeck) is designed to fuse feature information in complex logistics scenarios. GBForkDet significantly improves the capture of contextual logistics background information by reconstructing the Neck of YOLOv8 with BiODNeck. This is attributed to cross-layer weighted feature fusion and a complementary focus on learning convolutional kernels in any convolutional layer along all four dimensions of the kernel space. Furthermore, the introduction of the Normalized Wasserstein distance (NWD) as an enhanced loss function improves the detection of small distant objects in logistics scenarios. Experimental results show that GBForkDet achieves a mAP of 92.7% and 95.3% on the established Forklift-3k and KITTI datasets while reducing the model parameters by 17.9% and the computational cost by 22.5% compared to the baseline YOLOv8s model. Under the Jetson Nano edge platform and 640×640 input size, the GBForkDet model achieves a remarkable inference time of 108.2 ms using TensorRT acceleration.
13 August 2025 in Expert Systems with Applications
27 August 2025 in IEEE Transactions on Artificial Intelligence
Open World Object Detection (OWOD) aims to detect both known and unknown objects in dynamic environments, where unknown instances lack ground-truth supervision during training. Existing methods typically rely on supervision from known categories, leading models to overconfidently classify visually similar unknowns as known classes, and dissimilar ones as background. This known-class prior bias severely hinders the detection of truly novel objects. To address this challenge, we propose a robust unknown object detection method based on dual-granularity reconstruction error modeling. At the fine-grained level, we propose Fine-grained Masked Reconstruction (FMR), which randomly masks feature regions to guide reconstruction toward semantic structures, thereby improving foreground–background discrimination. At the coarse-grained level, we propose Adaptive Region-based Error Aggregation (AREA), which aggregates reconstruction errors over object proposals to enhance the model’s sensitivity to ambiguous semantic boundaries while suppressing local outliers. Furthermore, we perform decoupled probabilistic modeling of foreground and background reconstruction errors, enabling soft estimation of unknown object likelihoods without supervision. Extensive experiments on standard OWOD benchmarks demonstrate that our method consistently outperforms state-of-the-art approaches, achieving a +20.6 improvement in unknown object recall (U-Recall) while maintaining strong performance on known classes.