An Optimized Hybrid Soil Texture Classification Model using Stacked Sparse Autoencoder

• Published in: 2023 International Conference on Emerging Research in Computational Science (ICERCS) pp. 1 - 6
• Main Authors: Prabavathi, R, Chelliah, Balika J
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.12.2023
• Subjects: Clustering; Clustering algorithms; Computational modeling; Data analysis; Data models; Deep learning; Machine learning; Optimization; Parameter optimization; Scientific computing; Soil classification; Soil texture
• DOI: 10.1109/ICERCS57948.2023.10434248

With effective soil management, which lowers the costs associated with soil erosion, improves water utilisation, and other factors, soil data analysis can offer vital information to raise agricultural yield and nutrient utilisation efficiency. An unsupervised method of data analysis called clustering finds similar groups of objects by analysing attribute values. The dynamic subject of soil classification starts with the model and moves on to explain the classes and, at the end, apply the knowledge gained from the field. In light of this, the study presents a novel clustering and classification-based hybrid soil texture classification model (HSTC). Soil data will be clustered and categorised into multiple classes by the proposed (HSTC) model. Hydrological soil groupings (HSG) of various kinds are used as classes to accomplish this. Based on characteristics like the proportion of clay, silt, and sand, the HGS can be categorised. The density-based clustering (DBSCAN) algorithm is used to group the soil texture data. The stacked sparse auto encoder (SSAE) model is used to categorise the clustered data. In addition, the parameter values of the SSAE technique are determined using a new barnacle mating optimizer (BMO). The HSTC model has achieved the best results, with accuracy of 95.66%. The experimental validation of the proposed HSTC model on soil texture dataset revealed that the suggested model outperformed previous techniques.