Fractional Derivative Based TVD Smoothening and Baseline Correction for Extracting Leaf Wetness Duration From LW Sensor: A Novel Approach

One of the driving factors leading to the modernization in the agriculture sector is the era of sensors-driven technologies. Annually, as reported by the Associated Chambers of Commerce and Industry of India, $500 billion of crops are lost due to pests and plant diseases in a country like India, where at least 200 million Indians go to bed hungry every night. For the detection of plant disease, the measurement of leaf wetness duration (LWD) values becomes a crucial step. This requirement of measuring LWD values led to the development of an in situ IoT-enabled LW sensor earlier. The same LW sensor was deployed for about four months, and data for the same were collected. Furthermore, for extracting LWD information, smoothing algorithms like total variation denoising (TVD) are applied. However, our novelty lies in introducing the order of fractional derivative (α) in an already existing TVD algorithm, which is varied from 1 to 2, and results are found to be satisfying. To get an effective baseline, we combined this algorithm with three baseline correction techniques: asymmetric least squares, improved asymmetric least squares, and asymmetrically reweighted penalized least squares (arPLS). The optimal range of α lies in the range of 1.6 to 2 for getting the highest accuracy. This study demonstrates that our novel approach of integrating fractional derivatives into an existing TVD algorithm enhances its performance in identifying Leaf wetness events. The highest accuracy (i.e., the highest number of events detected) of 0.80 is found by total variation smoothing with the arPLS baseline correction technique.