Résumé: An Optimization Driven Spike Deep Belief Neural Networks is a type of neural network that is inspired by the functioning of the human brain. It is a variant of the more general class of Deep Belief Networks (DBNs), which are artificial neural networks composed of multiple layers of hidden units. Spike sorting is a critical process in neural signal processing that involves separating and identifying individual action potentials, spikes, from extracellular recordings of neuronal activity. This process is essential for understanding the behaviour of individual neurons and for decoding neural signals in various applications, such as Brain Machine Interfaces (BMIs) and neuro science research. Spike sorting is challenging due to the complexity of the recorded signals, including overlapping spikes and noise from other sources. This manuscript proposes A deep-learning based Multichannel Spike Sorting Neural Signal Processor (SSNSP) Module for High-Channel-Count Brain Machine Interfaces to record spike activity (SA) of brain neuron signals with less noise. Here first data acquisition is first step and the data's are took form Neural Signal Processor (NSP). Then the collected features are stored in BMIs. After this process feature is extracted using Haar DWT. Haar DWT is a frequency based feature extractor used to extract the spike or noisy signals from the neuron signals. Then the extracted features are given to driven spike DBN, this is a combination of multi-layer perceptron (MLP) layer and DBN. To increase the accuracy, Adam-Cuckoo Search optimization is used, which optimize the driven spike DBN weight parameter. An FPGA was used to construct and test a prototype 32-channel SSNSP component based on this analysis. Synthesised signals are used at various signals to noise ratios. Then, human neurons are classified based on the channels containing neural spike data. The impact of busy as well as idle state prediction errors on the spectrum efficacy is examined. The proposed technique is implemented in MATLAB platform. Finally, the proposed technique attains better detection accuracy 22.86%, 28.94%, 31.11% and 27.34% compared to the existing models, like Deep Learning Laser Speckle Contrast ESNN (DL-LSC-ESNN), Highly Stretchable Hydro gels as Wearable with Implantable Sensors for Recording Physiological with Brain Neural Signals (HSN-WIS-RPBNS), Lower-power band of neuronal spiking action dominated through local single units enhances the Presentation of BMI (LPB-NSA-LSU-BMI) and Emotion Categorization Utilizing Feature Fusion of Multimodal Data along Deep Learning in Brain-Inspired Spiking Neural Network (EC-FFMD-BISNN) respectively.

Langue: Anglais