Detection and identification of leaf miner and pest infestations in mangalore cucumber and wax gourd using YolovAM model in a complex environment

The adoption of computer vision techniques for disease detection is crucial, benefiting both technological advancement and the agricultural sector. Plant diseases and pest infestations significantly reduce crop yield and quality. The unpredictability of seasonal weather patterns, particularly due to frequent weather depressions, further complicates the farmer's ability to sustain the plant's growth and achieve optimal yield. For short-duration crops, timely monitoring of diseases is essential for effective crop management. Early and accurate disease detection reduces excessive pesticide use, thereby preserving soil fertility. This study contributes to the development of YolovAM, a custom object detection model designed to predict and identify leaf miner and pest infestation in Mangalore cucumber and Wax gourd. This is commonly caused by Liriomyza sativae, various beetle species, including cucumber beetles and red pumpkin beetles. The study was conducted using a dataset comprising 1670 images and nearly 10,230 granular annotations of Wax gourd and Mangalore cucumber plants. The dataset captured at various stages of leaf miner and pest infestation, ranging from early to advanced phases, across diverse farm conditions. The dataset presents a high level of complexity due to background noise, including soil, shadows, weeds, partially visible diseased leaf portions, and human interference. These factors pose significant challenges for conventional Yolo models in accurately detecting diseased areas, and YolovAM directly targets this problem. Experimental results indicate that YolovAM outperforms other object detection models by achieving 0.92 accuracy, 0.97 recall, 0.95 F1 score, and a mAP of 0.95.