Neural Network Based MATLAB Projects - MATLAB Solutions

Self-Organizing Map (SOM) neural network architecture for invariant facial feature similarity measurement, handling lighting variation and facial expressions.

• Problem & Objective: Map high-dimensional facial images onto 2D SOM feature topological grids for robust pattern recognition.
• Key MATLAB Functions: selforgmap, train, plotsomhits, plotsompos
• Expected Output/Metrics: Face recognition accuracy > 94%, SOM weight convergence plot, and confusion matrix under varying illumination.

Automatic isolated-word speech recognition using MFCC feature extraction and multi-layer feedforward back-propagation neural networks (BPNN).

• Problem & Objective: Extract Mel-Frequency Cepstral Coefficients (MFCC) from acoustic speech signals and classify spoken digit commands.
• Key MATLAB Functions: mfcc, feedforwardnet, train, confusionchart
• Expected Output/Metrics: Isolated word recognition accuracy > 96%, training MSE loss < 0.001, and acoustic feature spectro-temporal plots.

Extract time-frequency statistical features from power signals using Wavelet and S-Transform analysis to classify voltage sags, swells, transients, and harmonics with ANNs.

• Problem & Objective: Decompose non-stationary grid disturbances and classify multi-class power quality events.
• Key MATLAB Functions: modwt, st, patternnet, confusionmat
• Expected Output/Metrics: Disturbance event classification accuracy > 98.5%, multi-resolution wavelet energy plots, and noise robustness analysis.

Neuro-cryptography implementing Tree Parity Machines (TPM) for mutual neural network synchronization and secure secret key exchange over public channels.

• Problem & Objective: Synchronize weight vectors of sender and receiver neural networks without revealing encryption keys to man-in-the-middle attackers.
• Key MATLAB Functions: sign, hebbian, bitxor, plot
• Expected Output/Metrics: Synchronization time < 500 iterations, key space entropy > 7.99, and key agreement success rate = 100%.

Hermite Wavelet-embedded Neural Fuzzy (HWNF) adaptive Maximum Power Point Tracking (MPPT) controller for solar PV arrays under rapidly changing irradiance.

• Problem & Objective: Minimize MPPT tracking oscillations around maximum power point and accelerate dynamic response during cloud shading events.
• Key MATLAB Functions: anfis, genfis, sim, plot
• Expected Output/Metrics: Solar PV tracking efficiency > 99.2%, dynamic convergence time < 15 ms, and Total Harmonic Distortion (THD) < 2.5%.

Fast neural network approximator estimating traction power system voltage drops and substation loading constraints under varying railway train traffic intensities.

• Problem & Objective: Surrogate neural network modeling to replace slow iterative power flow solvers within railway investment planning optimization.
• Key MATLAB Functions: fitnet, trainlm, fmincon, perform
• Expected Output/Metrics: Computation speedup > 100x vs NR power flow, voltage drop prediction R² > 0.99, and capacity limit feasibility curves.

Biologically-inspired KIII chaotic neural network modeling mammalian olfactory cortex for volatile organic compound (VOC) gas identification from electronic nose sensors.

• Problem & Objective: Classify hazardous gas mixtures with high generalization capability under baseline sensor drift and ambient humidity fluctuations.
• Key MATLAB Functions: ode45, pca, fitcknn, confusionchart
• Expected Output/Metrics: VOC classification accuracy > 96.5%, 3D phase space trajectory plots, and sensor drift compensation performance.

Design Mamdani-style Fuzzy Logic Inference system for precise temperature regulation in thermal ovens, overriding nonlinear heat loss dynamics.

• Problem & Objective: Formulate membership functions and rule base for error and derivative error to maintain setpoint temperature without overshoot.
• Key MATLAB Functions: mamfis, addInput, addRule, evalfis, step
• Expected Output/Metrics: Settling time reduction (>30%), overshoot < 1.5%, and 3D fuzzy control surface plot.

Real-time KKT optimization and neural network energy management split strategy between high-density lithium batteries and high-power ultracapacitors in electric vehicles.

• Problem & Objective: Protect battery from peak transient currents during aggressive vehicle acceleration and regenerative braking.
• Key MATLAB Functions: fmincon, sim, legendre, plot
• Expected Output/Metrics: Battery peak RMS current reduction > 35%, battery degradation rate reduction (>25%), and ultracapacitor SOC recovery curves.

Face recognition system applying 2D Discrete Cosine Transform (2D-DCT) for dimensionality reduction followed by multi-layer perceptron neural network classification.

• Problem & Objective: Compress high-dimensional face pixels into low-frequency DCT energy coefficients for fast neural network training.
• Key MATLAB Functions: dct2, idct2, patternnet, train
• Expected Output/Metrics: Feature reduction ratio > 90%, classification accuracy > 95%, and execution time per image < 10 ms.

Implementation and harmonic spectral evaluation of Sinusoidal PWM (SPWM), Space Vector PWM (SVPWM), and Third-Harmonic Injection PWM (THIPWM) for 3-phase VSI inverters.

• Problem & Objective: Compare DC bus voltage utilization, switching losses, and Total Harmonic Distortion (THD) across three PWM modulation schemes.
• Key MATLAB Functions: power_fftscope, sim, thd, plot
• Expected Output/Metrics: SVPWM DC bus utilization gain (+15.47% vs SPWM), current THD < 3%, and inverter output harmonic spectrum plots.

Detect human faces in cluttered color images using YCbCr skin-color thresholding, morphological candidate region filtering, and Support Vector Machine (SVM) validation.

• Problem & Objective: Isolate face candidates under complex backgrounds and varying illumination conditions.
• Key MATLAB Functions: rgb2ycbcr, bwlabel, regionprops, fitcsvm
• Expected Output/Metrics: Face detection recall rate > 92%, false positive rate < 4%, and bounding box visual overlays.

Physiological stress detection using Skin Conductance (SC) galavanic sensor signals and finger temperature (FT) feature extraction with Random Forest classifiers.

• Problem & Objective: Extract tonic and phasic electrodermal activity (EDA) components to classify acute cognitive stress events.
• Key MATLAB Functions: cvx, findpeaks, fitcensemble, roc
• Expected Output/Metrics: Stress event classification accuracy > 91%, ROC curve AUC > 0.94, and physiological signal decomposition plots.

Fuzzy logic decision support controller evaluating feeder load imbalances and switching capacity to automatically reconfigure distribution network tie-switches.

• Problem & Objective: Minimize feeder overload risk and active line losses while keeping voltage profiles within 0.95-1.05 p.u. limits.
• Key MATLAB Functions: readfis, evalfis, power_loadflow, bar
• Expected Output/Metrics: Distribution loss reduction (>12%), feeder load variance reduction > 40%, and voltage profile plots before/after switching.

Automated farm security platform analyzing video feeds using optical flow and YOLO/CNN object detectors to identify unauthorized human and animal intrusions in real time.

• Problem & Objective: Detect multi-object perimeter breaches under low-light rural outdoor environments with automated alert triggers.
• Key MATLAB Functions: yolov4ObjectDetector, opticalFlowLK, detect, insertObjectAnnotation
• Expected Output/Metrics: Intrusion detection mAP > 88%, real-time video frame rate > 20 FPS, and zero false alerts from background foliage motion.