Deep Learning based Time Series Classification Model for Monitoring Land Cover Changes in Tropical High Forests

The monitoring of land cover/use in city areas is more important to have timely, accurate, and effective utilization and development of urban land resources.This remotely sensed dynamic monitoring greatly depends on the remote sensing data at high spatial and temporal resolutions.Still, the influence of weather and revisiting periods make time series images more complex.In this paper, a new land cover change detection model is implemented with three major three processes: Pre-processing, Detection of forest region, and Change detection of the forest.Initially, the input image is subjected to median filtering for outlier removal, reducing the influence of low-quality data and also filtering the noise.Once the image gets pre-processed, it is subjected to the detection of forest in the image by the optimized Convolutional Encoder-Decoder (OCED).The OCED classifier separates the forest region from the image as it resides with the cover of land, water, and forest as well.To make the process more effective, OCED gets enhanced by tuning the weight, and activation function are fine-tuned by the new algorithm termed as, Averaged Distance-based Sea Lion Optimization (ADSLnO) algorithm.Subsequently, the detected forest region is subjected to DTW-based time-series classification, which makes the year-wise change detection of the years 2007, 2009, 2015, and 2017.Finally, the performance of the adopted model is analyzed over other existing approaches in terms of certain metrics such as NPV, sensitivity, FDR, accuracy, specificity, FPR, precision, MCC, FNR, and F1-score, respectively.