In the contemporary landscape, the Internet of Things (IoT) has become a central focus across diverse sectors. Many IoT devices rely on machine learning (ML) models, endowing them with cognitive capabilities and decision-making proficiency. However, the inherent resource constraints of IoT devices often hinder the deployment of intricate ML models, such as deep learning (DL), on these edge devices. This research aims to enhance precision agriculture by employing methodologies that leverage the formidable capabilities of Deep Learning paradigms. The emphasis of the paper is on applying methods to enhance precision agriculture using deep learning on edge devices. In the proposed work, a Deep Neural Network (DNN) is trained and implemented to categorize images as valid or invalid in real time, reducing server load for subsequent tasks in precision agriculture, such as weed detection or crop identification. The DNN for classifying images will be implemented on an edge device, specifically a Microcontroller Unit (MCU). To harness deep learning capabilities on microcontrollers, the project will employ lightweight network modeling, quantizing parameters, and utilizing separable convolution to fit the model within the microcontroller's constraints. The model will be implemented on STM32 series microcontrollers for real-time inferencing. The NUCLEO-G474RE MCU kit and STM32CubeIDE were utilized for interfacing. The results underscore the potential of using deep learning models on edge devices for precision agriculture applications. The approach, involving separable convolution, data augmentation, batch normalization, dropout, and TFLite format conversion, proves effective in overcoming the challenges of limited computational resources on edge devices while achieving high accuracy and efficiency. The model demonstrated promising results on both test and real-world data, and the deployment process was simplified using STM32CubeIDE.