Enhancing Low-Resolution Images of Mustard Leaves Affected by Pests with Thermal Sensor using Super-Resolution Convolutional Neural Network Optimization

With urban areas facing limited agricultural land, hydroponic systems offer a solution to increase food storage and variety. Hydroponics, a farming technique that relies on water as a growing medium rather than soil, provides essential nutrients and oxygen for plants. This paper explores the use of thermal sensors to capture images of mustard leaves in a hydroponic system. In addition, it also explores thermal sensor images. These images are analyzed to detect pest attacks, with red leaves indicating the presence of pests and green/blue leaves unaffected by pests. These pests emit hot air; consequently, they turn red. The method of increasing resolution is to compare the Long Short-Term Memory (LSTM) algorithm with the Super-Resolution Convolutional Neural Network (SR-CNN) to improve the quality of images obtained from low-resolution sensors (AMG8833/Grid-EYE). The results show that the SR-CNN method is better than the LSTM (Long Short-Term Memory) method, although limitations remain due to the sensor resolution. After conducting the research, it could be observed that using LSTM resulted in a Mean Square Error (MSE) value of 0.001551685, while SR-CNN indicated an MSE value of 8.873. Furthermore, LSTM produces a Peak Signal-to-Noise Ratio (PSNR) value of 37.10797726, whereas SR-CNN achieves a PSNR of 39.199. The accuracy rates (SSIM) for LSTM and SR-CNN are 0.991538522961364 and 0.997747, respectively. These findings show that using the SR-CNN algorithm can effectively improve the quality of images produced by thermal sensors, even though the sensor pixel capacity is limited.