Bidirectional Long Short-Term Memory with Parametric Flatten-T Swish Activation Function for Land Use Land Cover Classification

The Land Use Land Cover (LULC) classification in aerial images involves analysing and extracting data to accurately classify various land cover types including urban areas, forests and agricultural fields.However, high dimensional feature space makes this task challenging, which leads to reduced classification performance.Therefore, Personal local search memory-based Woodpecker Mating Algorithm and Bidirectional Long Short-Term Memory with Parametric Flatten-T Swish (PWMA-BiLSTM with PFTS) are proposed in this research for LULC classification.The PWMA efficiently integrates global search capabilities with personal local search memory to enhance the exploration of the feature set.The BiLSTM captures complex and temporal patterns within the selected features through its bidirectional processing ability, which allows the capture of contextual data in LULC classification.The performance measures like accuracy, sensitivity, precision, f1-score, specificity and computational time are used to evaluate the performance of PWMA-BiLSTM with PFTS.The PWMA-BiLSTM with PFTS achieves an accuracy of 99.95% and 99.87% for Aerial Image (AID) and UC Merced (UCM) datasets, which outperforms Fully Convolutional Network (FCN).