The detection of fish freshness is crucial for ensuring food safety. This study addresses the limitations of traditional detection methods, which rely on laboratory equipment and complex procedures, by proposing a smartphone-based detection method, termed FreshFusionNet, that utilizes a pitaya peel pH intelligent indicator film in conjunction with multimodal deep learning. The pitaya peel indicator film, prepared using high-pressure homogenization technology, demonstrates a significant color change from dark red to yellow in response to the volatile alkaline substances released during fish spoilage. To construct a multimodal dataset, 3600 images of the indicator film were captured using a smartphone under various conditions (natural light and indoor light) and from multiple angles (0° to 120°), while simultaneously recording pH values, total volatile basic nitrogen (TVB-N), and total viable count (TVC) data. Based on the lightweight MobileNetV2 network, a Multi-scale Dilated Fusion Attention module (MDFA) was designed to enhance the robustness of color feature extraction. A Temporal Convolutional Network (TCN) was then used to model dynamic patterns in chemical indicators across spoilage stages, combined with a Context-Aware Gated Fusion (CAG-Fusion) mechanism to adaptively integrate image and chemical temporal features. Experimental results indicate that the overall classification accuracy of FreshFusionNet reaches 99.61%, with a single inference time of only 142 ± 40 milliseconds (tested on Xiaomi 14). This method eliminates the need for professional equipment and enables real-time, non-destructive detection of fish spoilage through smartphones, providing consumers and the food supply chain with a low-cost, portable quality-monitoring tool, thereby promoting the intelligent and universal development of food safety detection technology.
This study delves into the impact of artificial intelligence on students’ learning.The paper pioneers the use of k-means classification to reduce the dimensionality of features before the classification process, which minimizes redundant information and computational overhead. This step is not commonly found in prior research.Firstly, a random forest model is established with “whether AI assistance was used to complete the paper” as the output, and various features as inputs. The analysis of the predictive accuracy when changing input features reveals the degree of influence of each feature on using AI to complete the paper. After feature selection, the remaining 6 features can still effectively predict the use of AI to complete the paper, indicating the representativeness of the extracted indicators. Subsequently, the PCA-K-means algorithm is employed to classify the use of artificial intelligence. PCA algorithm is used for dimensionality reduction of the extracted features to better characterize students’ use of AI. Three representative variables are selected, and the contribution of each feature is computed based on the factor loading matrix. Finally, the K-means algorithm is used to classify students based on the three-dimensional features, showing that students can be effectively classified into 2 categories. The results indicate that the majority of students support the use of AI technology to enhance their abilities, although some students may only intend to use AI to cope with exams and similar tasks.
Rear-end collisions are highly related with car-following (CF) behaviors and account for a substantial proportion of road crashes. While existing countermeasures focus primarily on improving drivers' perception of the direct leading vehicle, growing evidence suggests that CF decisions are also influenced by information beyond the immediate leader. Thus, extending drivers’ perceptual range to include the indirect leading vehicle (i.e., the vehicle ahead of the direct leader) may improve CF safety. Compared with in-vehicle human machine interfaces (HMIs) that depend on vehicle to vehicle (V2V) communication, external HMIs (eHMIs) are more feasible at present. However, their effectiveness in improving CF safety under naturalistic driving environment remains unclear. To address this gap, three rear-facing eHMIs were designed: Brake-eHMI (braking action of indirect leading vehicle), Distance-eHMI and Headway-eHMI (relative distance and time headway between indirect leading vehicle and direct leading vehicle). An on-road experiment was conducted to evaluate the effect of eHMIs on CF safety. We found that, in general, indirect leading vehicle information could decrease the extreme safety-critical events. Among them, Distance-eHMI and Headway-eHMI also led to changes in CF strategies, as reflected in a reduced maximum desired acceleration. This study offers guidance for developing advanced vehicle systems that harness intelligent perception capabilities to improve driving safety.