A Scene Classification Method Based on Improved Incremental Brain-like Developmental Model

With the rapid development of machine learning, more and more vision-based scene classification methods are being applied to mobile robots to help them perform complex tasks such as their self-localization and navigation decisions. Most of today's scene classification algorithms are based on the batch learning approach, which often suffers from the inability to effectively process time-series data sets and models that are too complex and not scalable. To address the shortcomings and deficiencies of batch learning algorithms, this paper proposes an improved incremental brain-like developmental model, which can store the knowledge learned from the external environment in the neurons inside the model by simulating the activation of neurons in the human brain, and can learn and understand similar new knowledge based on the lobe component analysis (LCA). The model implements positive and negative feedback regulation mechanisms through the response competition mechanism of internal neurons. In this paper, the weight updating algorithm of the improved model can incrementally change the learning rate of the model and improve the accuracy of the scene classification task. Experimental results show that the model proposed in this paper has good recognition accuracy in indoor scene classification task. Compared with other traditional image classification algorithms for batch data processing, this model has better computational real-time performance by incremental learning of time-series data in the environment, which can meet the needs of mobile robots to handle complex scene classification tasks.