Table 1 Literature review

Kasinathan et al.³⁵

High classification rates of pest detection in Machine Learning techniques compared to traditional pest detection techniques. IP102 datasets used a pest detection algorithm that consists of foreground extraction and contour identification to detect pests

Convolutional Neural Network (CNN)

For 9 and 24 class insects the Maximum classification rate of 90% and 91.5% was achieved

Albanese et al.³⁶

Research focus was autonomous detection of pest infestation in orchard fruits. Regular pest detection and enhanced embedded low-power sensing system and neural accelerator capture and process the image in pheromone-based traps

Deep Neural Network (DNN)

Pest infestation detection task was automated for unlimited time and no farmer intervention was required

Suto et al.³⁷

Described the identification, detection, and counting of insects in agricultural pest management. A microcontroller board OpenMV Cam H7 which easily implicated the application using machine vision in real and embedded system-based insect trap

Deep Learning Based Insect Counting Algorithm

Pests counting and spraying methods were accurately scheduled

Chen et al.³⁸

A detecting drone used to take the image of the pest and recognize T, papillosa in the orchard tree and its position through YOLOv which was built on NVIDIA Jetson TX2 and embedded system

Tiny-YOLOv3 neural network model

Drone sprayed the pesticides on the exactly desired spots

Thenmozhi et al.³⁹

Focused on the classification of crop insects using machine vision and knowledge-based techniques. Features like texture, color, shape, histogram of oriented gradients (HOG), and global image descriptor (GIST) were used in the classification of pests. A 10-fold cross-validation test was conducted to achieve a better classification and identification of insects

Base classifiers (Naive Bayes, support vector machine and K-nearest-neighbor) and ensemble classifiers (random forest, bagging, and XGBoost)

Classification accuracy improved in the combination of texture, color, shape, HOG, and GIST features

Xiao et al.⁴⁰

Developed a pest identification network that combines deep learning and hyper spectral imaging technology for accurate pest control, paper discussed the limitations of traditional algorithms by utilizing one-dimensional convolution and attention mechanism to extract spectral features efficiently

Deep learning

Experimental results showed maximum suitability and accuracy for pest identification compared to other methods

Markovic et al.⁴¹

Monitored and predicted the presence of pest insects using cameras and sensor devices. Machine Learning models predicted the presence of pests which was based on temperature, humidity, and weather

ML algorithms

The affected detection accuracy was 86.3% and false detection percentage was 11%