Citation

Siddiquee, Kazy Noor e Alam (2025) Development of agriculture monitoring system implemented in Convolution Neural Networks with Simulated Annealing using Hybrid Energy Harvester. PhD thesis, Multimedia University.

Full text not available from this repository.

Abstract

Traditional Digital Agriculture Monitoring Systems (DAMS) were limited to using constrained supply of IoT sensor powers from bulky, costly and bio-hazardous electro-chemical cells. Moreover, market-aligned requirements of stakeholders have not been addressed through optimized stochastic solutions in developing algorithms for the food supply chain synchronised with sustainable development goals. To overcome such limitations, this research developed a DAMS with multiple algorithms such as detection and quantification, ripeness check and damage check for tomatoes and cucumbers. Again, the Feedforward Neural Network (FNN), Convolutional Neural Network (CNN), Faster CNN, Region based CNN (RCNN) & You Only Look Once (YOLOv5) methods were used for Machine Learning models; and Circular Hough Thresholding (CHT), Colour thresholding & Colour segmentation methods used for conventional image processing, have been incorporated both in MATLAB and Python. Their comparative performance analysis shows the best performances of CNN method achieving of 99.85% accuracy and costing of minimum simulation & training time among Jack-knife, K-fold 2, 5 and 10 cross-validations on diversified datasets. For individual datasets, the obtained error rates were 5.73% and 8.59% for tomatoes; and the error rates were 5.74% and 7.33% for cucumbers. For optimized stochastic solutions to DAMS, the articulation of sphere function and simultaneous equations in MATLAB compared the Genetic Algorithm (GA) with Simulated Annealing (SA) algorithm. The SA best performed compared to GA resulting with less generation of time and iterations. A new proposed CNN model which conceived with the selected SA was developed and the simulated in Python obtained the best accuracy of 87.28% with less iterations for meta-heuristics problems. Moreover, to avoid electro-chemical cells, a Hybrid Energy Harvester (HEH) was developed and designed using solar and vibration energy for driving an agriculture robot of the DAMS. The HEH consists of the GPS directed Maximum Power Point Tracking (MPPT) algorithm and its rectifier, boosters and supercapacitors were simulated in LTSPICE platform whereas they were designed and tested in the breadboard and implemented in the Printed Circuit Board (PCB) for verification with its LTSPICE results. The obtained voltages were 21.19V, 17.82V and 17.97V with a respective calculated power of 1.955W, 1.380W and 1.403W for LTSPICE simulation, breadboard and PCB circuit accordingly.

Item Type:	Thesis (PhD)
Additional Information:	Call No.: QA9.58 .K39 2025
Uncontrolled Keywords:	Algorithms
Subjects:	Q Science > QA Mathematics > QA1-43 General
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	19 Jan 2026 04:32
Last Modified:	19 Jan 2026 04:32
URII:	http://shdl.mmu.edu.my/id/eprint/15191

Downloads

Edit (login required)