Aiming at the problems of poor enhancement effect and large colour error in digital media image colour enhancement, a method of digital media image colour enhancement based on multi-scale Retinex is proposed. First, build and HSI colour space model to analyse digital media image colour space. Then, wavelet threshold method is used to deal with colour space noise. Finally, multi-scale Retinex is introduced to optimise Retinex through Gaussian filtering and weighted fusion to achieve colour enhancement of digital media images. The experimental results show that the proposed method can effectively improve the colour enhancement effect of digital media images, and the error is less than 0.15%, and the time cost is 0.20 s. This method effectively improves the colour enhancement effect.
Long English sentences in machine translation become the main factor affecting the quality of machine translation with its characteristics of more words and complex sentence structure. In this paper, we propose a translation method incorporating syntactic features in the field of natural language processing to address the problem of poor translation quality of English long sentences in machine translation. The method combines the linguistic template-based translation method and the conditional random field-based statistical translation method to slice and process long sentences by syntactic dimension and statistical dimension, and then improve the quality of machine translation. The experimental results show that the BLEU score of long sentences of English corpus improves about 2.1% and the NIST score improves about 1.1% after the training of the model proposed in this paper, which proves that the translation method of this paper improves the translation quality of long sentences of English.
Voice cloning has garnered significant attention for its ability to replicate individuals' voices using artificial intelligence. Existing methods include mel spectrogram and vector embedding approaches, each with strengths and weaknesses. This study introduces a hybrid pipeline leveraging both. It proposes a lightweight feed-forward neural network, crucial to the pipeline's performance. Unlike GAN-based architectures, this approach requires fewer data samples while achieving comparable results. The system uses a novel network as an encoder, Tacotron2 as a synthesiser, and WaveNet as a vocoder. The encoder captures distinctive vocal characteristics, generating speaker embeddings. Tacotron2 creates mel spectrograms for synthesised speech, and WaveNet produces high-quality audio waveforms resembling natural speech. The system is accessible in low-data scenarios, enabling faster training. Objective evaluations, including MOS, PESQ, and SNR metrics, confirm its superiority. This study presents a lightweight, data-efficient voice cloning system with applications in voice assistants, personalised speech synthesis, and entertainment.
The solution of differential equations finds many applications in a huge range of problems, and many techniques have been developed to approximate their solutions. For example, differential equations can be applied to physics problems, chemistry problems, economics, modelling, etc. This manuscript presents a number of global optimisation techniques that have been successfully applied to train machine learning models to approximate differential equation solutions. More specifically, two modified versions of genetic algorithms and particle swarm optimisation methods are proposed here. These methods have been successfully applied to solving ordinary differential equations and systems of differential equations as well as partial differential equations with Dirichlet boundary conditions.
Under the background of big data informationisation, the construction of multimodal education management mode has become a hot topic of discussion for efficient teaching reform. This paper combines the connotation of multimodal teaching, and establishes a teaching quality assessment model for college students based on big data technology. It focuses on the main advantages of multimodal teaching management and on how to use big data to predict the teaching quality of college students. This paper makes the network model meet the expectation by means of deep learning technology, and strengthens the multimodal teaching management model for college students. The research results provide some reference for improving the multimodal teaching management model of college students in the context of big data.
Travel review text reviews objectively reflect travellers' real perceptions of tourist destinations and services, and are also one of the important ways of internet word-of-mouth communication. According to the study, travellers will obtain information about products and other travellers' reviews of tourism through various channels before making purchase decisions, and use them as the basis for whether to continue purchasing. With the help of the research method of big data, the text takes the travel virtual community and the online review text of tourism on the web platform as the research material. To address the problems of previous models, we propose an attention-based mechanism LSTM, called SA-BiLSTM, for travel evaluation sentiment analysis. We integrate the attention mechanism into the LSTM and use it to improve the representation capability of the LSTM. The attention mechanism ignores the distance between words, which effectively solves the gradient disappearance and gradient explosion problems encountered by LSTM, and the combination of LSTM and attention mechanism is equivalent to model fusion at the structural level, which enables the model to capture information in different directions in the text and enhances the robustness of the model. We validate the good results of our model relative to state-of-the-art models on numerous real datasets.
Grammatical error correction aims to use computer programs to automatically correct grammatical errors in written texts. At present, the mainstream approach regards it as a monolingual translation task, and error correction is the process of translating 'wrong' sentences into 'correct' sentences. In this paper, we study the methods of error correction performance of ascending syntax from three aspects: model, training algorithm and data enhancement. The vast majority of syntax errors occur in a certain part of the text, but there are also a small number of syntax errors that span multiple segments of the text. We use transformer, the most advanced encoder decoder model based on attention mechanism in current neural machine translation, to model syntax error correction, so as to give consideration to local context information and long-distance dependency in text. The experimental results on two standard datasets show that transformer is significantly superior to models based on recurrent neural networks or convolutional neural networks.
Walking stability is an important index to measure the walking performance of robot, and is the key to realise its wide application. However, the traditional gait control system has low control accuracy and long control time. Therefore, aiming at the above problems, a nonlinear modelling and analysis method of stable behaviour of robot gait control system based on image processing technology is studied and designed. The robot gait control system is composed of main control system hardware, wireless control system, robot debugging software and image pre-processing module. The hardware design of the main control system includes the DSP minimum system and the steering gear control board. The image pre-processing module pre-processes the image collected by the CCD camera to remove the noise in the image. Finally, the speed of the robot is controlled by visual servo control method, and a nonlinear sliding mode control closed-loop system of the robot is constructed. The simulation results show that the designed method has the highest accuracy of 100%, and the longest time is only 27.96 s. It has high control accuracy and control efficiency, which provides a method reference for further realising the optimal control of the robot.
This study proposes a laboratory intelligent facial recognition system based on improved CNN, which significantly improves the accuracy of facial recognition by optimising the portrait recognition algorithm, improving CNN calculation and large parameter scale, and utilising perspective projection to improve portrait effect and sample utilisation. The experimental results show that the recognition rate has been improved by 10%, the CPU usage rate is less than 100%, and the model parameters have been reduced by more than 95%. This system can effectively and accurately recognise faces, making it suitable for embedded facial recognition devices.
In the visual processing of artistic images, traditional CNN has a high resource demand, and SNN can solve this problem. The article selects SNN as the method for artistic visual image processing and combines it with CNN to simplify model training. After CNN adjustment and feedback adjustment algorithm processing, the classification accuracy of SNN can be improved. The results show that the accuracy of the adjusted CNN model is 80.25% and 79.60%, respectively, with an average training accuracy difference of 3.32%. Under the same pooling combination, the accuracy of the model with 11 and 12 iterations is 68.00% and 66.02%, respectively. The average classification accuracy of SNN is 78.80%, slightly lower than the adjusted CNN. SNN has a power consumption of approximately 0.0039 W per second in processing 742 images. The correlation classification method used in the article can reduce power consumption and has a high classification accuracy.
