Underwater object detection is crucial in marine exploration, presenting a challenging problem in computer vision due to factors like light attenuation, scattering, and background interference. Existing underwater object detection models face challenges such as low robustness, extensive computation of model parameters, and a high false detection rate. To address these challenges, this paper proposes a lightweight underwater object detection method integrating deep learning and image enhancement. Firstly, FUnIE-GAN is employed to perform data enhancement to restore the authentic colors of underwater images, and subsequently, the restored images are fed into an enhanced object detection network named YOLOv7-GN proposed in this paper. Secondly, a lightweight higher-order attention layer aggregation network (ACC3-ELAN) is designed to improve the fusion perception of higher-order features in the backbone network. Moreover, the head network is enhanced by leveraging the interaction of multi-scale higher-order information, additionally fusing higher-order semantic information from features at different scales. To further streamline the entire network, we also introduce the AC-ELAN-t module, which is derived from pruning based on ACC3-ELAN. Finally, the algorithm undergoes practical testing on a biomimetic sea flatworm underwater robot. The experimental results on the DUO dataset show that our proposed method improves the performance of object detection in underwater environments. It provides a valuable reference for realizing object detection in underwater embedded devices with great practical potential.

Passion fruit, renowned for its significant nutritional, medicinal, and economic value, is extensively cultivated in subtropical regions such as China, India, and Vietnam. In the production and processing industry, the quality grading of passion fruit plays a crucial role in the supply chain. However, the current process relies heavily on manual labor, resulting in inefficiency and high costs, which reflects the importance of expanding the application of fruit appearance quality classification mechanisms based on computer vision. Moreover, the existing passion fruit detection algorithms mainly focus on real-time detection and overlook the quality-classification aspect. This paper proposes the ATC-YOLOv5 model based on deep learning for passion fruit detection and quality classification. First, an improved Asymptotic Feature Pyramid Network (APFN) is utilized as the feature-extraction network, which is the network modified in this study by adding weighted feature concat pathways. This optimization enhances the feature flow between different levels and nodes, allowing for the adaptive and asymptotic fusion of richer feature information related to passion fruit quality. Secondly, the Transformer Cross Stage Partial (TRCSP) layer is constructed based on the introduction of the Multi-Head Self-Attention (MHSA) layer in the Cross Stage Partial (CSP) layer, enabling the network to achieve a better performance in modeling long-range dependencies. In addition, the Coordinate Attention (CA) mechanism is introduced to enhance the network’s learning capacity for both local and non-local information, as well as the fine-grained features of passion fruit. Moreover, to validate the performance of the proposed model, a self-made passion fruit dataset is constructed to classify passion fruit into four quality grades. The original YOLOv5 serves as the baseline model. According to the experimental results, the mean average precision (mAP) of ATC-YOLOv5 reaches 95.36%, and the mean detection time (mDT) is 3.2 ms, which improves the mAP by 4.83% and the detection speed by 11.1%, and the number of parameters is reduced by 10.54% compared to the baseline, maintaining the lightweight characteristics while improving the accuracy. These experimental results validate the high detection efficiency of the proposed model for fruit quality classification, contributing to the realization of intelligent agriculture and fruit industries.

Accurate and prompt detection of cracked teeth plays a critical role for human oral health. The aim of this paper is to evaluate the performance of a tooth crack segmentation model (namely, FDB-DeepLabv3+) on optical microscopic images. The FDB-DeepLabv3+ model proposed here improves feature learning by replacing the backbone with ResNet50. Feature pyramid network (FPN) is introduced to fuse muti-level features. Densely linked atrous spatial pyramid pooling (Dense ASPP) is applied to achieve denser pixel sampling and wider receptive field. Bottleneck attention module (BAM) is embedded to enhance local feature extraction. Through testing on a self-made hidden cracked tooth dataset, the proposed method outperforms four classical networks (FCN, U-Net, SegNet, DeepLabv3+) on segmentation results in terms of mean pixel accuracy (MPA) and mean intersection over union (MIoU). The network achieves an increase of 11.41% in MPA and 12.14% in MIoU compared to DeepLabv3+. Ablation experiments shows that all the modifications are beneficial. An improved network is designed for segmenting tooth surface cracks with good overall performance and robustness, which may hold significant potential in computer-aided diagnosis of cracked teeth.

Dual-frequency modulation strategies in SSVEP-BCIs not only elicit fundamental responses but also induce robust intermodulation components (IMCs) through nonlinear neural integration, significantly expanding the available coding space. While IMCs can arise from various levels of the visual hierarchy depending on the stimulus complexity, the fundamental physiological origins of IMCs induced by pure luminance flicker—specifically whether they arise from retinal nonlinearity or cortical integration—remain debated. To resolve this, we employed a head-mounted display to physically isolate visual fields, comparing dichoptic hemifields (cortical) and monocular hemifield (retinal) conditions across varying spatial densities. Our results provide evidence that IMCs in scalp EEG during luminance modulation originate predominantly from retinal nonlinearities. Furthermore, while the geometric shape of the stimulus exerts a significant influence, the total length of the spatial boundaries separating the two frequencies acts as the dominant determinant of nonlinear output intensity. To elucidate why high-frequency IMCs often appear attenuated despite this robust boundary-driven generation, we constructed a spectral synthesis model (A = β·ITC·Freq^−α). This analysis demonstrated that the scalp-recorded amplitude is directly proportional to Inter-Trial Coherence (ITC) but is severely dampened by the brain’s intrinsic 1/f capacitive filtering, confirming that the apparent weakness of high-frequency components is an artifact of physical transmission.