Development of a Smartphone-Based Pest Identification System for Crop Management

Thesis Abstract

Crop pests significantly impact agricultural productivity worldwide, causing substantial economic losses and threatening food security, particularly in developing regions where access to expert pest identification services remains limited. Traditional pest identification methods rely heavily on manual inspection and expert judgment, which are often time-consuming, costly, and inaccessible to smallholder farmers. These challenges necessitate the development of efficient, cost-effective, and user-friendly technological solutions that leverage ubiquitous mobile devices to enhance pest management practices. The primary aim of this study was to develop and evaluate a Smartphone-Based Pest Identification System (SP-PIS) tailored for crop farmers to improve early pest detection and intervention. Specific objectives included (i) designing a mobile application integrated with an image processing and classification component; (ii) implementing a machine learning model, specifically a convolutional neural network (CNN), for pest recognition; (iii) assessing the system’s accuracy, usability, and acceptance among target users; and (iv) exploring the potential integration of the system into existing crop management frameworks. The research adopted a mixed-methods exploratory design combining quantitative and qualitative approaches. The quantitative component involved developing and training a CNN-based classification model utilizing a dataset of 10,000 pest images collected from agricultural research centers, local farms, and open-source repositories. For model validation, 2,000 labeled pest images not used in training were employed, and performance metrics such as accuracy, precision, recall, and F1-score were computed. The qualitative component involved semi-structured interviews and structured questionnaires administered to 150 smallholder farmers across three distinct agricultural zones, aiming to assess usability, perceived effectiveness, and willingness to adopt the system. Data collection instruments included a specially developed image dataset annotation protocol, a researcher-designed mobile application prototype, and structured questionnaires complemented by interview guides. Validity and reliability of quantitative instruments were ensured through cross-validation techniques, entropy-based data augmentation, and Cronbach’s alpha assessment of questionnaire internal consistency (?=0.86). Thematic analysis was employed to interpret qualitative data, while quantitative data analysis utilized statistical techniques including descriptive statistics, t-tests, and logistic regression analysis performed with SPSS and TensorFlow frameworks for model evaluation. The findings were anticipated to demonstrate an overall pest recognition accuracy exceeding 85%, with particular pest categories identified with over 90% precision. Usability testing expected to reflect high acceptance levels among farmers, with ease of use and perceived usefulness as key determinants influencing adoption intent. The system’s integration into crop management practices was projected to reduce pest identification time by up to 70%, thereby enabling timely pest control measures. The study’s innovative integration of mobile technology, machine learning, and participatory research provides a practical contribution to the digital transformation of pest management strategies for smallholder farmers. This research advances existing knowledge by demonstrating the feasibility of deploying real-time image-based pest identification tools in resource-constrained environments and offering a scalable model adaptable across diverse cropping systems. The study emphasizes the importance of user-centered design in ICT solutions and provides evidence for policymakers and agricultural extension services to promote digital literacy and technology adoption among farmers. Concluding, the study recommends the widespread dissemination of the developed system through extension services, further refinement of the algorithm for novel pest detection, and integration with weather and crop health monitoring platforms. Future research directions include expanding the pest database, incorporating multi-language support, and evaluating long-term impacts on crop yields and pest management efficiency. This comprehensive approach aims to enhance sustainable agricultural practices, contribute to digital agriculture, and ultimately bolster food security through technology-driven crop protection solutions.