Issuu

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 12 Issue: 10 | Oct 2025

p-ISSN: 2395-0072

www.irjet.net

AI-Driven Insect Recognition for Field Crop Management Yuktha N S1, Dr. Suresh D S2 1PG Scholar, Department of ECE, Channabasaveshwara Institute of Technology, Gubbi-572216, India 2Professor, , Department of ECE, Channabasaveshwara Institute of Technology, Gubbi-572216, India

---------------------------------------------------------------------***--------------------------------------------------------------------significant threats to rice plants. The Brown Planthopper Abstract - Leveraging artificial intelligence (AI), we propose an advanced insect recognition system designed to revolutionize field crop management and boost sustainable pest control. This initiative employs deep learning and computer vision techniques, utilizing models extensively trained on diverse agricultural pest datasets, to achieve rapid and precise identification of insect species from on-site images or live video. The primary goal is to deliver real-time field surveillance and early detection capabilities, empowering farmers to implement prompt, targeted pest management strategies. This approach minimizes the use of conventional, broad-spectrum chemical treatments, effectively reduces crop losses, and fosters both increased productivity and ecological responsibility in agriculture.

does damage by feeding on phloem sap and by vectoring viral infections like ragged stunt and grassy stunt. Its infestations can induce an incidence of "hopper burn" such that the entire plant turns brown and dies prematurely. The Rice Ear Bug, which occurs during the period of grain filling, feeds on ripening grains and produces unfilled or discoloured grains, thereby affecting the quality of grains. The Rice Leaf Folder larvae roll the leaves and feeds inside, masking the plant's photosynthetic potential and causing losses of yield. The Rice Stem Borer, probably the most feared pest, enters the stem and cuts off the supply of nutrients, producing dead hearts (in seedlings) and whiteheads (in mature plantations).

Key Words: CNNs-Convolutional Neural Network, Deep

1.1 Problem Statement

Learning, CNN Architecture, Train-Val Split.

Accurate and timely detection of insect pests is critical to successful integrated pest management of rice crops. At present, however, integrated pest management relies heavily on the field experts' eye, through training, which is laborious and subject to human failure. Since most pest species, such as Brown Planthopper, Rice Ear Bug, Rice Leaf Folder, and Rice Stem Borer, are morphologically related, even seasoned experts are likely to mistake them, more so under harsh field situations. Additional environmental elements such as changing light sources, overlapping of insects, and even natural cover hinder the visualization of insects and lead to late or incorrect responses, thus causing extensive damage to the crops.

1.INTRODUCTION Agriculture contributes significantly to food security, livelihood, and economy in most parts of the globe, especially in developing nations. Of the various crops, rice serves as a primary staple food to more than half of the global population. Its productivity is, however, adversely affected by various insect pests attacking at more than a single development stage, from seedling to harvesting stages. Visual assessment and human recognition by skilled entomologists or field specialists have constituted the traditional means of pest management, but they are timeconsuming and may be ineffective and subjective, even in commercial production. The incorporation of prevailing technology trends, more so artificial intelligence (AI), and machine learning (ML), offers potential solutions to the autonomous recognition and categorization of insect pests under agricultural scenarios.

In addition, traditional machine learning approaches employed in pest classification rely significantly on handcrafted features and tend to perform poorly in practical field situations with sophisticated backgrounds. Furthermore, large, balanced, and annotated datasets of specific rice pests are few, with still fewer resources accessible for less-studied species such as the Rice Ear Bug. Furthermore, a majority of these deep learning models require costly computational laboratories, limiting their use in field applications where hand-held or embedded devices are of choice. For this reason, creating an efficient, semi-automated system of classification consisting of CNN, which are in a position to learn from realistic agricultural observations and give accurate, online pest recognition straight from images acquired in the field, continues to be of utmost significance.

Over the past few years, DL approaches, and more importantly, CNNs have shown outstanding effectiveness recognition of data patternsand image classification applications. These approaches have the capacity to learn autonomouslycomplex visual patterns from images without the significance for hand-crafting of features. For the task of pest detection, CNNs have the potential to distinguish insect species from subtle dissimilarity of shape, colour, texture, and posture. The proposed system is aiming to implement a system of classifying pests using CNNs and train them to learn and recognize four key pests of rice. All the pests are

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 406