This paper proposes a lightweight and efficient mango detection model named Light-YOLO based on the Darknet53 structure, aiming to rapidly and accurately detect mango fruits in natural environments, effectively mitigating instances of false or missed detection. We incorporate the bidirectional connection module and skip connection module into the Darknet53 structure and compressed the number of channels of the neck, which minimizes the number of parameters and FLOPs. Moreover, we integrate structural heavy parameter technology into C2f, redesign the Bottleneck based on the principles of the residual structure, and introduce an EMA attention mechanism to amplify the network’s emphasis on pivotal features. Lastly, the Downsampling Block within the backbone network is modified, transitioning it from the CBS Block to a Multi-branch–Large-Kernel Downsampling Block. This modification aims to enhance the network’s receptive field, thereby further improving its detection performance. Based on the experimental results, it achieves a noteworthy mAP of 64.0% and an impressive mAP0.5 of 96.1% on the ACFR Mango dataset with parameters and FLOPs at only 1.96 M and 3.65 G. In comparison to advanced target detection models like YOLOv5, YOLOv6, YOLOv7, and YOLOv8, it achieves improved detection outcomes while utilizing fewer parameters and FLOPs.

Object detection is one of the most critical areas in computer vision, and it plays an essential role in a variety of practice scenarios. However, small object detection has always been a key and difficult problem in the field of object detection. Therefore, considering the balance between the effectiveness and efficiency of the small object detection algorithm, this study proposes an improved YOLOX detection algorithm (BGD‐YOLOX) to improve the detection effect of small objects. We present the BigGhost module, which combines the Ghost model with a modulated deformable convolution to optimize the YOLOX for greater accuracy. At the same time, it can reduce the inference time by reducing the number of parameters and the amount of computation. The experimental results show that BGD‐YOLOX has a higher average accuracy rate in terms of small target detection, with mAP0.5 up to 88.3% and mAP0.95 up to 56.7%, which surpasses the most advanced object detection algorithms such as EfficientDet, CenterNet, and YOLOv4.

The presence of non-tobacco-related materials can significantly compromise the quality of tobacco. To accurately detect non-tobacco-related materials, this study introduces a lightweight and real-time detection model derived from the YOLOv11 framework, named LRNTRM-YOLO. Initially, due to the sub-optimal accuracy in detecting diminutive non-tobacco-related materials, the model was augmented by incorporating an additional layer dedicated to enhancing the detection of small targets, thereby improving the overall accuracy. Furthermore, an attention mechanism was incorporated into the backbone network to focus on the features of the detection targets, thereby improving the detection efficacy of the model. Simultaneously, for the introduction of the SIoU loss function, the angular vector between the bounding box regressions was utilized to define the loss function, thus improving the training efficiency of the model. Following these enhancements, a channel pruning technique was employed to streamline the network, which not only reduced the parameter count but also expedited the inference process, yielding a more compact model for non-tobacco-related material detection. The experimental results on the NTRM dataset indicate that the LRNTRM-YOLO model achieved a mean average precision (mAP) of 92.9%, surpassing the baseline model by a margin of 4.8%. Additionally, there was a 68.3% reduction in the parameters and a 15.9% decrease in floating-point operations compared to the baseline model. Comparative analysis with prominent models confirmed the superiority of the proposed model in terms of its lightweight architecture, high accuracy, and real-time capabilities, thereby offering an innovative and practical solution for detecting non-tobacco-related materials in the future.

Infrared and visible image fusion (IVIF) aims to integrate complementary information from infrared and visible images, combining thermal target saliency with texture and structural details. Despite significant progress, existing deep fusion methods still do not fully exploit stage-specific complementary cues, and their higher-stage relation modeling may become unreliable when semantic-context guidance and structural reliability constraints are limited. To address these issues, we propose a Multi-Granularity Interaction and Semantic-Structure Rectification Network, termed MISSFusion. MISSFusion focuses on stage-aware integration of mid-level multi-granularity complementary interaction and higher-stage semantic-structure rectification. Specifically, MISSFusion introduces a Multi-Granularity Complementary Interaction Block (MCIB) to balance local details and regional context while exploiting consistency and discrepancy cues between infrared and visible features at mid-level stages. MISSFusion further develops a Semantic-Structure Rectification Attention (SSRA) module to rectify higher-stage feature relations under semantic-context guidance and structural reliability constraints, thereby improving structural consistency and suppressing unreliable relation propagation. A joint objective is designed to preserve complementary information while regularizing semantic-structure rectification. Extensive experiments on multiple IVIF benchmarks and downstream tasks suggest that MISSFusion achieves competitive performance on most evaluation metrics while maintaining a balanced trade-off among target saliency, texture preservation, and structural consistency.