“Public Acceptance” Rating Model of Classical Music

Yikang Hong — Fri, 26 Sep 2025 00:00:00 +0800

A large proportion of classical music is overlooked by the general public but has the potential to become widely accepted. This paper proposes a model to rate the “public acceptance” score, ranging from 0 to 100, of an arbitrary music piece. It serves as an objective, convenient, and effective way of musical analysis. The model involves a standard dataset for comparison, a preprocessing module, a feature extractor, and a rating module. The standard dataset consists of music pieces that have already gained “public acceptance” according to statistics, while the resampling and normalization module preprocesses the audio, and the feature extractor uses pre-trained VGGish to extract vector features of the audio. Then, these features are compared with those in the standard dataset, and the final rating is calculated. In addition, the model has features that make the rating explainable, and it has been proven by objective and subjective evaluation that this model is accurate and reliable, with the ratings on the same piece highly converged and close to that of manual rating. Possible improvements to this model are marked in the end.

Optimization and OpenCV-Based Implementation of Face Detection Systems

Jian Sun, Yizheng Xu — Sun, 28 Sep 2025 00:00:00 +0800

Face detection, as a fundamental task in computer vision, holds significant value in applications such as security surveillance, human-computer interaction, and identity recognition. OpenCV, as an open-source computer vision library, provides various efficient face detection algorithms; however, it still faces challenges in complex scenarios, including insufficient detection accuracy and poor adaptability. This study systematically investigates the performance of traditional cascade classifiers (Haar and LBP) and deep learning models (DNN module) within the OpenCV framework, proposing a series of improvement methods.Firstly, through experimentation, we analyze the impact of detection parameters (e.g., scale factor, minimum neighbors) on performance and optimize the baseline detection pipeline. Secondly, to address challenges like illumination variations and pose diversity, we propose enhancement strategies based on image preprocessing (histogram equalization, noise suppression) and post-processing (optimized non-maximum suppression, false detection filtering). Furthermore, we explore a hybrid detection approach that combines cascade classifiers with deep learning models to improve robustness. Experimental results demonstrate that the improved method significantly enhances detection precision and recall rates on FDDB and WIDER FACE datasets, particularly showing better adaptability to low-light conditions, occlusions, and multi-angle faces. This research provides practical solutions for optimizing face detection in OpenCV environments, offering valuable references for related application development.

Intelligent Systems & Robotic Mechanics

“Public Acceptance” Rating Model of Classical Music

Optimization and OpenCV-Based Implementation of Face Detection Systems