Implementation of Convolutional Neural Network for Soil Type Category Detection in a Web-Based Plant Recommendation System

Abstract

The growth of the agricultural sector in Indonesia is highly dependent on soil fertility, as soil is an important factor in the agricultural sector. However, conventional identification of soil types often takes a long time and requires high costs. To overcome this problem, this research develops a soil classification system using an optimized Convolutional Neural Network (CNN) model to improve soil classification accuracy. The results of this classification become the basis for a Content-Based Filtering (CBF) based recommendation system, in order to provide suggestions for crop types that are suitable for soil types. This research was conducted through several main stages, namely soil image data collection, data preprocessing, CNN model training and CBF-based recommendation system implementation. The CNN model is used to recognize soil texture and color patterns, while CBF is used to match soil characteristics with suitable plant species. System evaluation is conducted using confusion matrix to assess the accuracy of the classification model as well as the effectiveness of the recommendation system. The soil type classification process using CNN with MobileNetV2 architecture achieved an accuracy rate of 96%. This result shows that the architecture is effective in recognizing soil types precisely and can be used to provide appropriate crop recommendations. Thus, this system has the potential to support increased agricultural productivity, both on a small and large scale.