Smart Agriculture: Machine Learning And Deep Learning For Advanced Weed Detection And Crop Quality Enhancement

The evolution of agricultural systems is ongoing, driven by new technologies that seek to refine traditional farming. The primary goal is to not only boost agricultural output per hectare but also to improve crop quality, all while maintaining the natural environment. Notably, weeds present a considerable threat to crops, as they deplete essential nutrients, water, and light, thus diminishing crop productivity. The uniform spraying of entire fields for weed control results in high costs and detrimental environmental impacts. To mitigate the limitations of standard weed control methods, this research proposes the use of Machine Learning (ML) and Deep Learning (DL) methods to identify and classify weeds within crops. For ML-based approaches, various statistical and texture-based features are extracted, including image moments, mean absolute deviation, and gray level co-occurrence matrix (GLCM). The YOLOv8m algorithm is utilized for weed identification, and for weed classification, features from the CottonWeedID15 and Earlycrop-weed datasets are used to train SVM, Random Forest, and ANN models, using SMOTE to balance the classes. Deep learning models, such as VGG and ConvNeXtBase, are also trained on balanced data for automated feature extraction and classification. The best ML result was a 99.5% accuracy with SVM, and the best DL result was 98% accuracy with ConvNeXt and Random Forest. This demonstrates the potential of these methods for efficient agricultural solutions