The detection of students’ behaviors in classroom can provide a guideline for assessing the effectiveness of classroom teaching. This study proposes a classroom behavior detection algorithm using an improved object detection model (i.e., YOLOv5). First, the feature pyramid structure (FPN+PAN) in the neck network of the original YOLOv5 model is combined with a weighted bidirectional feature pyramid network (BiFPN). They are subsequently processed with feature fusion of different scales of the object to mine the fine-grained features of different behaviors. Second, a spatial and channel convolutional attention mechanism (CBAM) is added between the neck network and the prediction network to make the model focus on the object information to improve the detection accuracy. Finally, the original non-maximum suppression is improved using the distance-based intersection ratio (DIoU) to improve the discrimination of occluded objects. A series of experiments were conducted on our new established dataset which includes four types of behaviors: listening, looking down, lying down, and standing. The results demonstrated that the algorithm proposed in this study can accurately detect various student behaviors, and the accuracy was higher than that of the YOLOv5 model. By comparing the effects of student behavior detection in different scenarios, the improved algorithm had an average accuracy of 89.8% and a recall of 90.4%, both of which were better than the compared detection algorithms.

In the context of non-independent and identically distributed (non-IID) data and limited client participation, repeated iterations of local models in federated learning worsen model drift, resulting in directional bias during global aggregation. Existing single-end optimization approaches, such as client-side regularization or server-side control variables, have difficulty maintaining global-local consistency. To address this, we introduce the Federated Learning with Dual-end Gradient Correction and Proxy-free Self-Distillation (FedDGC-PSD) algorithm. On the client side, a self-distillation mechanism based on historical models constrains local prediction consistency via a Kullback-Leibler (KL) divergence loss, mitigating overfitting without the need for external data; concurrently, a gradient correction term aligns local updates with the global direction using historical gradients. On the server, dynamic cross-round gradient correction terms use weighted historical gradients to reduce variance accumulation. Extensive tests were run on five datasets against four benchmark algorithms. Results show FedDGC-PSD outperforms FedAvg, FedProx, SCAFFOLD, and FedDyn by 13.41%, 14.05%, 20.23%, and 3.41%, respectively, on CIFAR-100. In a large-scale scenario with 1,000 clients and 1.5% participation, FedDGC-PSD maintains stable convergence on CIFAR-10, outperforming FedDyn by 17.62%. Ablation studies show that combining the Dual-end Gradient Correction (DGC) and Proxy-free Self-Distillation (PSD) modules improves global model accuracy by about 0.71%, 2.26%, 4.50%, and 6.16% on MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100, respectively, when compared to using only the DGC module.

Underwater robots are an important tool for exploiting marine resources, while underwater target detection is a key technology for marine fishing by underwater robots. For the problem that the traditional algorithm is difficult to detect small targets and multiple scales in the underwater environment, this paper proposes an underwater biological detection algorithm based on the multi-scale feature fusion and attention mechanism. First, a YOLOv5 network is constructed, and a bi-directionally weighted BiFPN feature pyramid structure is fed into the neck network so that multiscale target features can be efficiently merged. Second, a channel and spatial convolutional attention mechanism is introduced between the neck network and the prediction network in order to increase the saliency of the underwater target that needs to be detected, which effectively reduces the missed and erroneous detection of underwater targets. The experimental results show that the proposed algorithm can accurately detect multi-scale underwater targets in complex underwater environments with an average accuracy of 80.18%, which is 2.87% better than YOLOv5, and an average recall of 85.06%, which is 4.81% better than YOLOv5.