Improved Hybrid CNN Bidirectional Long Short-Term Memory with BIRCH based segmentation for Crop and Weed Classification

Weed is an undesired plant that grows along with all field crops including rice, wheat, soybeans, sugar beet, cassava, and millet. It can affect crop yield and quality. Identifying and classifying weeds can indeed be challenging for farmers due to their similarities and the potential negative impact they can have on crop growth. Pre-processing, segmentation, extraction of features, and classification are the four processing steps for categorizing crops and weeds. In this work, a BIRCH algorithm is introduced for the segmenting process along with a deep learning model for the classification process. The aim of the work was to categorize weeds and crop species based on various features, including the MBP and GLCM, are extracted from the segmented outcome. Finally, the hybrid model classification system, which is a combination of improved Bi-LSTM and CNN, receives the obtained features as input and classifies whether they are weeds or crops. The hybrid model determines the classification of crops and weeds by averaging the output of two classifiers, which gives 97.33% accuracy.