| 摘要: | 隨著時代的進步，科技對於人們的影響越來越深，不論是經濟、交通、生活品質或是娛樂上都少不了它的蹤影，蔬菜水果的品質的要求也越來越高，使得農業與科技正在逐步靠近。人們可以利用專業技術去達成環境控制、品質分析、自動摘採等，使農產達到更高品質或是生產上有更高的效率。
「智慧農業」是人類對於科技與農業結合的目標，當中的自動採收也是熱門的研究之一，自動採收少不了偵測目標物的辨識系統。過去傳統方法的辨識系統對於環境有著很高的要求，環境會直接性的影響到影像偵測或分析的成功率，因此大多數的實驗者會於控制的環境之下進行研究。AI影像學習是目前影像辨識的主流之一，它是人工智慧當中的一個分支，利用大數據去訓練系統，讓系統可以判別目標物的特徵，使不同環境下的偵測物都可以被考慮進去，因此大幅降低了環境對於辨識的影響。
本研究使用機器學習技術，擷取芭樂於農場環境之位置及座標，並利用色彩分析遮罩等處理進行防撞泡棉去除及腐壞區域偵測。為了提高腐壞辨識分析的成功率，使用了白平衡、物件去背、中值濾波或是空間顏色量化等方法，將雜訊去除和色彩還原，來提高圖像中芭樂的防撞泡棉及腐壞特徵。
本實驗用400張芭樂圖片進行測驗，前置芭樂位置辨識成功率為61.34%，後段色彩分析遮罩之成功率為79.47%，故本實驗對於芭樂之位置及腐壞辨識分析成功率為48.75%。;With the advancement of technology, its influence on people is becoming more and more profound. Whether it is in the areas of economy, transportation, quality of life, or entertainment, its presence is ubiquitous. The demand for high-quality fruits and vegetables is also increasing, which is causing agriculture and technology to gradually come closer together. With the use of professional techniques such as environmental control, quality analysis, and automatic harvesting, agricultural products can achieve higher quality and production efficiency.
"Smart agriculture" is the goal of humans to combine technology and agriculture, and automatic harvesting is one of the popular research areas. Automatic harvesting requires an object recognition system to detect the target. Traditional recognition systems in the past had high environmental requirements, and the environment would directly affect the success rate of image detection or analysis. Therefore, most researchers conducted studies in a controlled environment. AI image learning is currently one of the mainstreams of image recognition, and it is a branch of artificial intelligence that uses big data to train the system to recognize the characteristics of the target object. This enables the detection of objects in different environments to be considered, thereby significantly reducing the impact of the environment on recognition. Artificial intelligence is the mainstream of this generation and will also be an important chapter in human technology.
In this research, Image machine learning techniques were used to capture the location and coordinates of guava within the farm environment. Color analysis, masking, and other processing methods were employed to remove collision foam and detect areas of decay. To enhance the success rate of decay recognition analysis, methods such as white balance, object extraction, median filtering, and spatial color quantization were utilized to remove noise and restore colors, thus improving the identification of collision foam and decay features in the guava images.
In this experiment, a total of 400 guava images were used for testing. The success rate of the initial guava location identification was 61.34%. The subsequent color analysis masking achieved a success rate of 79.47%. Therefore, the overall success rate of guava position and decay recognition analysis in this experiment was 48.75%. |
